<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Dom on Medium]]></title>
        <description><![CDATA[Stories by Dom on Medium]]></description>
        <link>https://medium.com/@DomTheAssistant?source=rss-1a84e29b81aa------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*_KAnqjZ7EbII35kuUlEszg.jpeg</url>
            <title>Stories by Dom on Medium</title>
            <link>https://medium.com/@DomTheAssistant?source=rss-1a84e29b81aa------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 15:50:40 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@DomTheAssistant/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[USP (TR-369)]]></title>
            <link>https://medium.com/@DomTheAssistant/usp-tr-369-db59eff08430?source=rss-1a84e29b81aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/db59eff08430</guid>
            <category><![CDATA[bbf]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[connected-home]]></category>
            <category><![CDATA[usp]]></category>
            <category><![CDATA[internet-of-things]]></category>
            <dc:creator><![CDATA[Dom]]></dc:creator>
            <pubDate>Fri, 30 Aug 2019 09:56:32 GMT</pubDate>
            <atom:updated>2019-08-30T11:03:14.859Z</atom:updated>
            <content:encoded><![CDATA[<h4>Why ISPs should care and why we care.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zw0XgBKADV6zkq8gQN30Hg.png" /></figure><p>It’s been difficult for ISPs to offer new services or technology via the home gateway (CPE). The challenges in developing, testing and managing new services on gateways with variations in hardware, software and firmware make it difficult for any party to deploy services. Managing them over time is also difficult and this creates a lot of risk for the service providers. It also makes it very difficult to test, iterate and upgrade the services frequently.</p><h3>How USP brings flexibility to deploying and updating services</h3><p>The User Services Platform (USP, TR-369) is the next generation of TR-069. USP enables service providers, vendors and end-users to manage, monitor, upgrade and control connected devices and services. This is done through an “Agent” and “Controller” relationship that enables remote manipulation of software and hardware capabilities. An agent can expose its capabilities to one or more controllers, allowing an agent to be controlled by multiple controllers and a controller can control multiple agents.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hau8vzqut7xFuxSs2r8ekA.png" /><figcaption>Example of a controller/agent relationship in a connected home</figcaption></figure><p>USP is developed by the Broadband Forum as an open standard and is gaining a lot of traction from vendors and providers. It uses the device:2 data model which can also be found in TR-069, it can live side-by-side with legacy devices as well as new devices. By setting a standard for how to remotely manage services on the gateway it creates flexibility for all parties to update frequently without upgrading the entire firmware or software.</p><h3>Why does Domos care?</h3><p>USP makes it easier for us to deploy and manage our machine learning technology on the gateways. Which means we can update our models frequently and perform real-time actions and learn from them. Machine Learning is dependent on rapid upgrade cycles and the ability to learn by trial-and-error. USP enables us, in a safe manner, to learn and adapt quickly to every connected home and their unique environment.</p><p>For more information about USP, visit <a href="https://usp.technology/">https://usp.technology/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=db59eff08430" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Domos Update Q2 2019]]></title>
            <link>https://medium.com/@DomTheAssistant/domos-update-q2-2019-de9ff3267ce3?source=rss-1a84e29b81aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/de9ff3267ce3</guid>
            <category><![CDATA[ufbb]]></category>
            <category><![CDATA[qed]]></category>
            <category><![CDATA[usp]]></category>
            <category><![CDATA[bbf]]></category>
            <category><![CDATA[wifi]]></category>
            <dc:creator><![CDATA[Dom]]></dc:creator>
            <pubDate>Thu, 06 Jun 2019 10:34:45 GMT</pubDate>
            <atom:updated>2019-06-06T11:09:47.337Z</atom:updated>
            <content:encoded><![CDATA[<h4>USP (TR-369) white paper — QED, a true metric for internet Quality of Experience — Wi-Fi “Fast Track” for online gaming — Domos Forum.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*qZXiKba5FXoUrpq-.png" /></figure><h3>Gaming, USP &amp; Knowledge</h3><p>In this edition we will bring you news about the new USP white paper co-authored by us containing use cases and examples, how our Wi-Fi “Fast Track” technology can improve online gaming, Broadband QED and the launch of our Domos Forum.</p><p>Spring is here, finally!<br><em>Hugs &amp; high fives, your friendly Home IT Assistant</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/772/0*1QO_LNg-NziTerpu.png" /></figure><h4>Controlling the Connected Home</h4><p><em>“Machine Learning offers a new opportunity with the potential of automating management and optimization of home networks. Machine Learning methodologies can find the optimal configuration for every single device. To enable this, vast amounts of data need to be aggregated and gathered in a cloud solution.</em>”</p><p>More about this and the benefits of BroadBand Forums USP in the <br><a href="https://www.broadband-forum.org/wp-content/uploads/2019/05/BBF-179-TR-369-MU-461-Marketing-Update-V6.pdf"><strong>newly released paper</strong></a> that has been co-authored with Magnus Olden, our Head of Data Science.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*vhIuh1YkxEcLu7r9.jpg" /></figure><p>We are attending this year UFBB BASe event in Den Haag and our Head of Data Science will be presenting on day 3. Hope to see you there! <br>For more information about the event, <a href="https://www.broadband-forum.org/meetings-and-events/basecamp-ultra-fast-broadband">click here</a>.</p><h3>Wi-Fi “Fast Track” for Gaming</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ln3Wy0XAvEcsrbTPL2cX_g.png" /></figure><p>Gaming is the most demanding use-case for a typical home network. To deliver a good gaming experience the network must provide low and stable latency, minimal packet loss, and consistent throughput. Guaranteeing this level of quality requires a way of allocating resources to clear a path for the gaming traffic in the home network.</p><p><a href="https://medium.com/domos-creating-the-home-it-assistant/wi-fi-fast-track-to-make-gaming-first-138cac17ea58">Wi-Fi Fast Track to make Gaming First</a></p><h3>Broadband QED</h3><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FYy5_SpDpRF8%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DYy5_SpDpRF8&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FYy5_SpDpRF8%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/307d5c9a4a8fbf5a4cf2a931523c2ae9/href">https://medium.com/media/307d5c9a4a8fbf5a4cf2a931523c2ae9/href</a></iframe><p>Led by Vodafone and Predictable Network Solutions (PNSol), this initiative brings together one of Broadband Forum’s oldest members with the newest member Domos to create “invisible” networks that will greatly enhance the quality of experience broadband networks provide.<br><a href="https://www.broadband-forum.org/neil-davies-pnsol-and-magnus-olden-domos-discuss-the-broadband-qed-project">Click here</a><a href="https://www.broadband-forum.org/open-broadband/broadband-qed"> for more info.</a></p><h3>Domos Forum</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dtqoN004igMbRPOt59bhzQ.png" /></figure><p>We collaborate and share our insights and data in multiple industry and data science networks. This makes Domos a great hub for sharing and developing data science expertise in an open forum. That is why we are bringing our knowledge and insight to companies and events that will benefit from our learnings over the years.</p><p><a href="https://medium.com/domos-creating-the-home-it-assistant/domos-forum-a3e34fd058bc">Domos Forum</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=de9ff3267ce3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Wi-Fi Fast Track to make Gaming First]]></title>
            <link>https://medium.com/@DomTheAssistant/wi-fi-fast-track-to-make-gaming-first-138cac17ea58?source=rss-1a84e29b81aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/138cac17ea58</guid>
            <category><![CDATA[algorithms]]></category>
            <category><![CDATA[network]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[gaming]]></category>
            <category><![CDATA[wifi]]></category>
            <dc:creator><![CDATA[Dom]]></dc:creator>
            <pubDate>Wed, 05 Jun 2019 11:45:34 GMT</pubDate>
            <atom:updated>2019-06-05T11:45:34.426Z</atom:updated>
            <content:encoded><![CDATA[<h4>Guaranteeing a perfect gaming experience by optimising individual application outcomes</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ln3Wy0XAvEcsrbTPL2cX_g.png" /></figure><p>Gaming is the most demanding use-case for a typical home network. To deliver a good gaming experience the network must provide low and stable latency, minimal packet loss, and consistent throughput. Guaranteeing this level of quality requires a way of allocating resources to clear a path for the gaming traffic in the home network.</p><p>At Domos we have developed a Wi-Fi slicing technology we call <strong>Wi-Fi Fast Track. </strong>It is an advanced machine learning algorithm designed to optimise individual application outcomes through real-time service classification and dynamic allocation/slicing of Wi-Fi network resources. What the Wi-Fi Fast Track technology does is to guarantee the delivery of specific services like gaming, streaming and video calls. We know that one of the big market potentials for this technology is in the gaming segment in the connected home. With a high sensitivity to latency and a large growth in both adaptation and networking requirements for the past few years, we see gaming as a great place to start deploying this algorithm.</p><p>To better understand the gaming market we initiated a survey in collaboration with one of our customers. The survey focused on how their customers were gaming and if they would be positive to a service that improves gaming over <br>Wi-Fi. The key numbers from the survey are shown below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o93ZVgGhtDog2ZsHjZ5_LQ.png" /></figure><p><strong>The customers want stable gaming over Wi-Fi and they are willing to pay a decent monthly fee for it as well</strong>. This is a big market potential not only in terms of new revenues but also in customer loyalty, satisfaction, marketing content and increased sales. Another finding in the survey was the response rate they received from the customers targeted using our Device Taxonomy data vs random selection of customers. <strong>The response rate was twice as high for the target group using our data</strong>. That’s a great benefit for both parties, as our customer received more responses, but customers not in the target group also received less non-relevant information. In this survey, our customer used Device Taxonomy to identify customers who had gaming devices and equipment such as Nintendo Switch, Xbox One, Oculus VR, Playstation 4 and much more. Based on the survey we also got a good indication of how many customers are gaming on which platforms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2dus8zs55-S0gfIZ_DrYoA.png" /></figure><p>Based on this insight we started collaborating with several ISPs, partners and vendors in order to make this a reality for gamers. <strong>We are currently planning a market soft launch with a leading Nordic operator this year, in collaboration with a top tier OEM vendor.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=138cac17ea58" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Domos Forum]]></title>
            <link>https://medium.com/@DomTheAssistant/domos-forum-a3e34fd058bc?source=rss-1a84e29b81aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/a3e34fd058bc</guid>
            <category><![CDATA[internet]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[wifi]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Dom]]></dc:creator>
            <pubDate>Fri, 31 May 2019 15:01:43 GMT</pubDate>
            <atom:updated>2019-07-30T11:05:24.355Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>We bring our data and experience to you for a live talk.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x6iFhg1HOAzJHuMYWlxcag.png" /><figcaption>Magnus Olden presenting at the Data Science Meetup in StartupLab 29th of January, 2019.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8NeW_bT2APqVulfD0lCe1Q.jpeg" /><figcaption>Bjørn Ivar Teigen presenting at the Fu5ion Plenary Autumn Meeting in Rennes 4th of October, 2018.</figcaption></figure><p>Domos has a team of data science experts with years of experience working directly with data from the home and data models to properly develop machine learning algorithms. Our data science experts are working with data sets of unprecedented quality from 100,000 homes - logging data continuously on a second-by-second basis.</p><p>We collaborate and share our insights and data in multiple industry and data science networks. This makes Domos a great hub for sharing and developing data science expertise in an open forum. That is why we are bringing our knowledge and insight to companies and events that will benefit from our learnings over the years. This article contains the talks we currently provide and will be updated as we learn and understand even more.</p><p>For more information about how to collaborate with us or how to get us to your company/event, please reach out to us on info@domos.no</p><h3>Service Classification &amp; Quality of Experience</h3><p>Service classification is the process of identifying services running on a network by looking at traffic patterns. Domos has developed a system for this which can detect specific services such as Fortnite and Skype. Combined with a new and better way to measure network quality, called QED, this technology allows ISPs to optimise for application outcomes. QED is a framework for measuring and reasoning about network performance which is now in the process of standardization through the broadband forum (BBF). The technique has successfully been used to improve network performance at CERN’s ATLAS experiment. It has also been used to measure the impact of network upgrades and to detect network problems by several major European ISPs with Vodafone as one of the main adopters.</p><p>In this talk, Bjørn Ivar Teigen will be presenting the two processes and why we think they are critical in order to achieve correct use of network resources based on the requirements each application needs to deliver excellent Quality of Experience to the users.</p><h3>Mimicking Evolution For Better Wi-Fi In Dense Areas</h3><p>Everyone hates unstable, unreliable Wi-Fi, and still, it’s everywhere. One of the most common reasons for slow Wi-Fi is other routers. There is a limited amount of data that can be transmitted through the radio spectrum at any time, and other routers on the same channel as yours will reduce your precious download and upload speeds.</p><p>Competing with other routers for the same capacity is known as Wi-Fi congestion. In large apartment complexes, there may be hundreds of routers that share the same frequency as there are only 3 non-overlapping channels to choose from on the widely used 2.4GHz band.</p><p>In densely populated areas, Wi-Fi often breaks down because it is highly congested. In this talk, Bjørn Ivar presents the challenges associated with congestion and a way to improve Wi-Fi in populated areas. At Domos we utilize our cloud solution along with genetic algorithms, to evolve the best channel assignments for each case. This reduces the number of competing neighbours and improves the utilization of the radio resources. The solution is tried and tested through actual deployment, and we showcase real-world results.</p><p><strong>About the speaker</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/947/0*qorYoqFKRlAJcoQZ" /><figcaption>Bjørn Ivar Teigen, Researcher &amp; Data Scientist at Domos</figcaption></figure><p>Bjørn Ivar Teigen is a researcher and data scientist at Domos with a Master’s Degree in Robotics and Intelligent Systems from The University of Oslo. He is currently doing a PhD co-sponsored by Domos and The Research Council of Norway titled “<em>Hierarchical Reinforcement Learning Models with Applications in Radio Resource Management</em>” in the field of Machine Learning and Computer Networks. He has previously worked on and developed several algorithms at Domos related to optimisation of networks in congested areas, classifying traffic based on packet sizes and other research initiatives we are involved in. To read more about what Domos is working on including articles written by Bjørn Ivar, please visit our Medium Publication here: <a href="https://medium.com/domos-creating-the-home-it-assistant">https://medium.com/domos-creating-the-home-it-assistant</a>.</p><p><a href="https://twitter.com/bjornivart?s=17">Bjørn Ivar Teigen (@bjornivart) | Twitter</a></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FmlARMsQXzCs%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DmlARMsQXzCs&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FmlARMsQXzCs%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/515b92b7104f2a3678e0b5d3d7fcc9a0/href">https://medium.com/media/515b92b7104f2a3678e0b5d3d7fcc9a0/href</a></iframe><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FIOhOg1NC4T4%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DIOhOg1NC4T4&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FIOhOg1NC4T4%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/55be39fcfb1b5a689fe5388d5ab9b80d/href">https://medium.com/media/55be39fcfb1b5a689fe5388d5ab9b80d/href</a></iframe><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FOKl31ncYLSw%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DOKl31ncYLSw&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FOKl31ncYLSw%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/a0a62b1550b7700b46010b303f9eff07/href">https://medium.com/media/a0a62b1550b7700b46010b303f9eff07/href</a></iframe><h3>What’s Wrong With Wi-Fi?</h3><p>Domos have over the last year measured Wi-Fi Quality of Experience metrics in over 100 000 homes on a second by second basis. Through this data gathering and extensive research, we’ve learned to identify the cause and effects of low QoE.</p><p>The major pain is data transmitted slowly, or with low data rates as is industry jargon. We show that the primary cause of low data rates is not poor signal-strength as is popular belief, but end-user devices’ suboptimal selection of data rates. We show that the data rates set by devices are significantly lower than those set by the routers and that device types and device vendors show distinct behavioural patterns. And that it should be possible, though complex, to significantly increase speeds through configurations on the router.</p><p>Understanding Wi-Fi QoE is the core of Domos’ research, and we invite others to join us in attempts to optimize it on current and future hardware and protocols. Our preliminary results show that with machine learning and statistical techniques, optimization is possible by recognizing device types. We can learn to manipulate the behaviour of their rate adaptation algorithm, finding their strengths and weaknesses in different types of environments and for different types of usage.</p><p>This talk has been previously held for: Oslo Data Science Meetup, Telenor Research and TNOs Ultrafast Broadband seminar in Den Haag (now UFBB BASe).</p><h3>Enabling ISP Data-Monetization in a Post-GDPR World</h3><p>Our lives are increasingly digital. For the time we spend in our homes, every digital activity and experience can be tracked in the internet gateway (CPE). Better insight into this data can help the ISPs deliver higher quality products and get an exceptional understanding regarding users. Unfortunately, much of this data is sensitive and it is hard to argue that the ISP needs to store this data, an important part of GDPR. Additionally, collecting and deriving insight from the data in scale is challenging.</p><p>In this presentation, Magnus Olden explains how an ISP could achieve high levels of customer insight in a privacy-sensitive way. This can be done by using a Federated Learning approach and/or Secure Aggregation Protocols to create non-reversible, self-constraining methods for learning about users. This will facilitate happier customers and the ability to identify the groups of customers that are most likely to churn/cross and upsell/give high NPS scores, etc. All within the reach of GDPR.</p><p>He describes how deep insight can enable new business lines and revenue streams. The way Amazon knows the patterns of what you shop, Google what you do on the internet and Facebook about your social life — this presentation shows how the same approach can give the ISP the knowledge for where we spend most of our time: At home.</p><p><strong>About the speaker</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*oZAeM5yQ_DOEsb2Q" /><figcaption>Magnus Olden, Head of Data Science at Domos</figcaption></figure><p>Magnus Olden is the Head of Data Science at Domos with a Master’s Degree in Informatics: Robotics and Intelligent Systems from the University of Oslo. He is specialized in Wi-Fi, CPE software stack and Machine Learning, currently involved in several projects and initiatives within these domains. He is an active member of the Connected Home Council in the Broadband Forum and co-writer of the USP (TR-369) white paper published by the Broadband Forum. To read more about what Domos is working on including articles written by Magnus Olden, please visit our Medium Publication here: <a href="https://medium.com/domos-creating-the-home-it-assistant">https://medium.com/domos-creating-the-home-it-assistant</a></p><p><a href="https://twitter.com/OldenMagnus?s=17">Magnus Olden (@OldenMagnus) | Twitter</a></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FM0oZlSHMQMA%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DM0oZlSHMQMA&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FM0oZlSHMQMA%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/f7abf309de58403577b197105d023efe/href">https://medium.com/media/f7abf309de58403577b197105d023efe/href</a></iframe><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FqEQ6ye5r6zw%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DqEQ6ye5r6zw&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FqEQ6ye5r6zw%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/88be3ce416f3ca7f9aed17d38af410d4/href">https://medium.com/media/88be3ce416f3ca7f9aed17d38af410d4/href</a></iframe><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FZC_t97gxvdY%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DZC_t97gxvdY&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FZC_t97gxvdY%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/0c065c1bf9622603095758654b47d961/href">https://medium.com/media/0c065c1bf9622603095758654b47d961/href</a></iframe><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FYy5_SpDpRF8%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DYy5_SpDpRF8&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FYy5_SpDpRF8%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/307d5c9a4a8fbf5a4cf2a931523c2ae9/href">https://medium.com/media/307d5c9a4a8fbf5a4cf2a931523c2ae9/href</a></iframe><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2Fvideoseries%3Flist%3DPLLp8LhUlrW4Ey5-2xHSpwImliTusD44uI%26start%3D1&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DEPFUw0O4G0w&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FEPFUw0O4G0w%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/ea9646731bc2d7ab70aba1dcf0ed7913/href">https://medium.com/media/ea9646731bc2d7ab70aba1dcf0ed7913/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a3e34fd058bc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Google’s Stadia was just announced, but nobody is talking about the Wi-Fi management required for…]]></title>
            <link>https://medium.com/@DomTheAssistant/googles-stadia-was-just-announced-but-nobody-is-talking-about-the-wi-fi-management-required-for-9fffa87b1e49?source=rss-1a84e29b81aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/9fffa87b1e49</guid>
            <category><![CDATA[google]]></category>
            <category><![CDATA[wifi]]></category>
            <category><![CDATA[gaming]]></category>
            <category><![CDATA[stadia]]></category>
            <dc:creator><![CDATA[Dom]]></dc:creator>
            <pubDate>Mon, 01 Apr 2019 09:49:38 GMT</pubDate>
            <atom:updated>2019-04-01T09:49:38.428Z</atom:updated>
            <content:encoded><![CDATA[<h3>Google’s Stadia was just announced, but nobody is talking about the Wi-Fi management required for it to work.</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/0*ntHEYzoltvg7UBY7.jpg" /><figcaption>Credit: REUTERS/Stephen Lam</figcaption></figure><p>Google just announced Stadia, their seamless cloud gaming platform promising online gaming experience on every screen. Playstation and Xbox have already launched their own services offering streaming-based games. That means you can play without downloading the games to the actual consoles. This is achieved by running the game in the cloud. There are several services offering a similar experience to PC gamers. It will bring gaming to a completely new level of engagement, interaction across screens and services as well as new ways of using this technological advancement for gaming development.</p><p>Everyone is talking about the requirements on the provider’s side, what type of infrastructure they need to deliver, and how users can access the content. The problem nobody is talking about is the resources required on the edge to actually stream games over Wi-Fi in the home. Expectations will be low-latency and stability of the streamed content to keep a good quality of experience. Stadia boss Phil Harrison said in an <a href="https://kotaku.com/google-stadia-boss-answers-and-dodges-our-questions-1833422137">interview with Kotaku</a> that in order to run Stadia on 1080p, 60 frames per second, the player requires 25mbps and 30mbps to run it in 4K. If the bandwidth is lower, they will turn down the resolution.</p><p>But how does this look like over Wi-Fi? Imagine how today’s video conference services are working over Wi-Fi. For those of us using it on a regular basis, either for business or personal reasons, it’s known to be quite unstable, changing in quality and breaking down more than we’d like. A video call requires about 1–2Mbps. Now multiply that with 20x and you have both the expectations and failed quality delivered for cloud gaming with today’s Wi-Fi Quality of Service. It will result in bad service delivery for Google and bad and unstable user experience for all the gamers.</p><p>The next question is if this is addressable today, and if so what is required to deliver 1080p or 4K gaming quality over cloud gaming to customers on Wi-Fi. The answer is Wi-Fi slicing, and this is why ISPs need to start implementing it in order to deliver the service quality needed for products like Stadia.</p><h3>Wi-Fi Slicing to ensure Quality of Experience</h3><p>If Google’s Stadia is to deliver a good gaming experience on Wi-Fi, which is the premise of the service, allocating networking resources for devices using it will be absolutely necessary. Wi-Fi slicing is the method of allocating specific networking resources to services used in the home based on available resources and what each service requires for a good Quality of Experience. This varies from service to service, but for Stadia the most important thing will be throughput stability. Secondly is latency, which will need to be minimised all the way from Wi-Fi to the server and back.</p><p>As Stadia can be consumed on a range of different devices along with other services at the same time, detecting the Stadia traffic is vital for Wi-Fi slicing to work. At Domos we have developed a unique machine learning algorithm called <strong>Service Classification</strong> that can detect and classify all types of services used in the home over Wi-Fi. By using Service Classification to detect Stadia traffic, the gaming flow can be allocated throughput resources needed for 1080p or 4K resolution without interruptions in quality.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9fffa87b1e49" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Nordic Innovators Cloudstreet and Domos Partner to Deliver an End-to-End, Application-Aware 5G…]]></title>
            <link>https://medium.com/@DomTheAssistant/nordic-innovators-cloudstreet-and-domos-partner-to-deliver-an-end-to-end-application-aware-5g-dc01dfc2fb8c?source=rss-1a84e29b81aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/dc01dfc2fb8c</guid>
            <category><![CDATA[internet-of-things]]></category>
            <category><![CDATA[5g]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[network-slicing]]></category>
            <category><![CDATA[telecom]]></category>
            <dc:creator><![CDATA[Dom]]></dc:creator>
            <pubDate>Fri, 22 Feb 2019 10:54:02 GMT</pubDate>
            <atom:updated>2019-02-22T13:44:43.853Z</atom:updated>
            <content:encoded><![CDATA[<h3>Nordic Innovators Cloudstreet and Domos Partner to Deliver an End-to-End, Application-Aware 5G Experience to the Home</h3><p><em>Cloudstreet’s Carrier-grade API and Network Slicing Engine with Domos’ Machine Learning For the Home Solution delivers intelligent connectivity right out of the box.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*7UWEPQ_ISGGRzLC9oEoQvQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*GHHnxWzJSCxZdpdi.png" /></figure><p>Barcelona, Spain., Feb. 22, 2019 (GLOBE NEWSWIRE) — Finland’s<a href="https://www.globenewswire.com/Tracker?data=VhZpRHIsvgCAHBmPCZCJBX_zvt4U6dCiTX3zS2CXJpYm9io73gOGJCu5mzV4XrUaGetcRa2PUPNKpaTwFb-EWQ=="><strong> Cloudstreet</strong></a>, the US-patented Network Slicing Company, and Norway’s <a href="https://www.globenewswire.com/Tracker?data=q1dpneeKmft1QAtL65Mse9xfKayMI0ryFzLWs2jpIM69lKjSVt2QU1NBM2SiE9Tl"><strong>Domos</strong></a>, a leader in Machine Learning technologies for the smart home, are pleased to announce that they have teamed up to create the industry’s first end-to-end, intelligent application and context-aware network slicing solution for home networks. The solution will be on display February 25–28 at the Mobile World Congress in Barcelona in Hall 5, stand 5C41.</p><p>A perfect example of MWC’s aspirational theme, “Intelligent Connectivity”, the solution closes the loop on delivering a 5G experience to fixed wireless home networks.</p><p>The combined technologies solve two key, last-mile problems that have plagued mobile-enabled home networks:</p><p>1) How to build application-awareness into the network without compromising performance and compliance (i.e. DPI)</p><p>2) How to manage and orchestrate QoS-assured slices triggered from User Equipment (UE) across serving nodes and base stations and out onto the Internet.</p><p>This unique combination of Dynamic RAN QoS Management &amp; Orchestration, provided by the Cloudstreet <a href="https://www.globenewswire.com/Tracker?data=4XXmNjUd8ew4iiNYpK3WrzbvzHCgTiiKBBZPdhrA4znZ8_22jy02QeF1SpzpxUcSa-c0FM-xa2f1MEKXXbO0bvSfbEWnVcyhTkahyonrEyU6I3VD4TY0syMHtviI8D0k"><strong>Dynamic Profile Controller™ (DPC™)</strong></a>, with <a href="https://www.globenewswire.com/Tracker?data=JVVHSfBJ6IX2UC9hDQuIRnfmu0pjOoJod_Aab3626C6Rri6G_yS8urO4o7dhD3U4o94RMPUiW0XZ8H54_BVz6JZSQhF73O78lnhazxWDpnA="><strong>Domos’ home CPE</strong></a> installed non-DPI app-sniffing, machine learning algorithm solves these problems, delivering real-time context-optimized QoE-guaranteed connectivity.</p><blockquote>“This is not about Network Slicing, 5G, capacity, latency or throughput. No consumers will ever know or care about those things,” said Cloudstreet Founder and CEO Mika Skarp. “This is about delivering digital experiences that are markedly better than they were before, and consistently, reliably so. We’re thrilled to be working with Domos, a company who’s CPE-centered machine learning technology so perfectly complements our own”.</blockquote><blockquote>“Our focus has always been machine learning technology that secures Quality-of-Experience within the home,” said Domos Co-Founder and CTO Olav Nedrelid “Teaming up with Cloudstreet, we can finally secure these diverse services across mobile and fixed wireless networks for an array of services including Online Gaming, OTT streaming and Video Calls with dynamic Quality-of-Experience from End-to-End. Digital experiences in the home are notoriously unreliable. With our combined technologies, we can, for the first time offer SLA-assured, machine-informed connectivity over in-home Wi-Fi networks” he said.</blockquote><p>Together, Cloudstreet and Domos are delivering on the promise of intelligent connectivity and the expectation of 5G networks well in advance of their large-scale arrival. The companies look forward to the opportunity to present the combined solution to interested carriers at MWC19, Hall 5 Stand 5C41 and throughout the year at their convenience.</p><p><strong>About Cloudstreet</strong><br>Finland-based Cloudstreet is a US-patented Network Slicing innovator dedicated to revolutionizing the mobile experience, empowering user choice and driving new MNO revenues. Providing a platform for Intelligent, Application-Aware Connectivity and Dynamic (Reactive and Predictive QoS), Cloudstreet allows carriers to offer on-demand, SLA-assured slices via its telecom-grade APIs and award-winning Dynamic Profile Controller™ engine. Tailoring Quality-of-Experience to user demand and context, Cloudstreet today delivers a fleet of mission-critical use cases from Fixed Wireless to Broadcasting to Public Safety. Awarded the EC’s Horizon 2020 grant, TMForum’s Catalyst Award and Best Connectivity Solution (WCA16), Cloudstreet’s in-market solutions are deployable in today’s 4G and emerging 5G networks. Visit us at <a href="https://www.globenewswire.com/Tracker?data=Tvvp_BJBsynTMFy7zoWEnZvWZf7Gptd26fHSWeNtD555e7ZTaeTtVXDT_fWyKclZLm7QJ-UKnQ1XRs49UjIkZdkawh1M4bm3Amz74TcLDgk="><strong>www.cloudstreet.co</strong></a>.</p><p><strong>About Domos</strong><br>Norway-based Domos provide patent-pending machine learning algorithms that assure optimal Quality-of-Experience for all digital services consumed in the home. Domos’ solution leverage existing CPEs to provide actionable insight that optimises and stabilise in-home Wi-Fi, all through automation and machine learning algorithms. This reduces direct cost in customer care, churn and marketing spend but also create exciting new revenue opportunities that have the potential of transforming the carriers business potential. For more information, contact us at <a href="https://www.globenewswire.com/Tracker?data=9wssErNeuimbwEOzxiQU7cGkJJqB7ZpOa2Q5sob0kEk0ciWmfGkVdy5PCLrK-ecqOXCu7VnOiFus5iEgws7x0w=="><strong>sales@domos.no</strong></a> or visit our website at <a href="https://www.globenewswire.com/Tracker?data=AmB6mbxZgVLnPQbL9WGXRdF3z7lmFGEr5IT3jBCZMGWSLdcdG0o0UjS3JlySE5M4R_GTbrOCBZfw83rz1Cvn3g=="><strong>www.domos.no</strong></a></p><h4>MEDIA CONTACTS:</h4><p><strong>Brendan Tully Walsh</strong></p><p>Marketing Head, Cloudstreet</p><p>brendan.walsh@cloudstreet.co</p><p>+1 (514)-802–6099</p><p><strong>Odin Ottem Berntsen</strong></p><p>Head of Communications &amp; Customer Experience</p><p>odin@domos.no</p><p>+47 97 98 96 68</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dc01dfc2fb8c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is Wi-Fi EasyMesh?]]></title>
            <link>https://medium.com/@DomTheAssistant/what-is-wi-fi-easymesh-b48c8c4160c7?source=rss-1a84e29b81aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/b48c8c4160c7</guid>
            <category><![CDATA[mesh-networks]]></category>
            <category><![CDATA[wifi]]></category>
            <category><![CDATA[internet-of-things]]></category>
            <category><![CDATA[easymesh]]></category>
            <category><![CDATA[home]]></category>
            <dc:creator><![CDATA[Dom]]></dc:creator>
            <pubDate>Tue, 22 May 2018 09:11:42 GMT</pubDate>
            <atom:updated>2018-06-05T14:22:11.754Z</atom:updated>
            <content:encoded><![CDATA[<h4>Wi-Fi Alliance announce EasyMesh — Turning every Wi-Fi router into a Mesh node.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*uMid-x7LEUWQv92x." /><figcaption>Credit: <a href="https://www.wi-fi.org/discover-wi-fi/wi-fi-easymesh">https://www.wi-fi.org/discover-wi-fi/wi-fi-easymesh</a></figcaption></figure><h3>What is Wi-Fi EasyMesh?</h3><p><a href="https://www.wi-fi.org/discover-wi-fi/wi-fi-easymesh">Wi-Fi EasyMesh</a> is a new standard from the Wi-Fi Alliance for Wi-Fi roaming and optimisation across multiple access points (i.e. “Wi-Fi mesh” or “Whole Home Wi-Fi”). The standard is supported by all major chipset makers, including Broadcom and Qualcomm. It is backward compatible to 802.11n, meaning that most current CPEs could become EasyMesh compatible through a software upgrade.</p><h3>Why it matters</h3><p>The new standard will create interoperability and flexibility to operators. They can leverage existing devices through software upgrade, and they can later add hardware from any vendor that is EasyMesh compatible. As the standard has support from all major chipset makers, it is likely to quickly commoditise the emerging, but proprietary Whole Home Wi-Fi market. It also marks the entry of Broadcom, which until now was notably absent from the space.</p><h3>Background</h3><p>It has only been two years since Silicon Valley startup Eero <a href="https://www.engadget.com/2016/02/23/eero-is-the-home-wifi-solution-ive-been-waiting-for/">launched their home Wi-Fi “solution”</a>, kicking off the Whole Home Wi-Fi solution market that now counts dozens of vendors and was expected to grow to double digit billion US$ by 2025.</p><p>Now, these growth predictions were perhaps already in doubt as reports of layoffs and acqui-hires started appearing in the tech press. Still, operators were sensing the raised expectations (and willingness to pay) in their customer base and continued their testing of the retail offerings. Many offered to sell them directly to their customers, but a general lack of device management support left many of these initiatives half-hearted. Also, Broadcom was notably missing from the space as most of the retail-focused Wi-Fi mesh solutions were based on Qualcomm Atheros chipsets.</p><p>Vendors like AirTies and Plume have actively been addressing the operator market, but seemed to already be transitioning from hardware based offerings to software that could adapt to third party hardware. In recent months, Nokia and FonTech were adding similar offerings through acquisitions.</p><h3>What will happen next?</h3><p>We believe operators will support this new standard 100%, and push EasyMesh software upgrade to their largely Broadcom-based device portfolios as soon as these upgrades are available. Given the sheer volume of that installed base, this will likely force the other chipset makers to follow suit and offer similar software upgrades of legacy operator provided equipment. With standard-based interoperability in place, operators can support and distribute a range of devices from multiple vendors. We believe many of the legacy retail vendors will choose to support the standard to access this market opportunity. It will also be very hard for market leader Google Wi-Fi not to support the standard, with their bigger picture strategic goals and reputation risk.</p><p>This leaves startups like Eero, Plume and others, who will likely be commoditised. Hardware will be plug-and-play, and the “intelligent” software management will be built into the protocol. The Wi-Fi Alliance will once again have succeeded to enabling the multi-vendor interoperability and common feature set which has been the foundation of Wi-Fi’s success.</p><p>Of course, Mesh can only solve coverage issues. As recent research clearly demonstrate, <a href="https://medium.com/domos-creating-the-home-it-assistant/whats-wrong-with-wi-fi-f29a84f3eac6">the dominant problem today is poor transfer rates</a>, especially from the client devices. Hopefully the launch of EasyMesh can increase operators investment in <a href="http://domos.no/products.html">solutions addressing these issues</a>.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2Fd8f5c0%3Fas_embed%3Dtrue&amp;dntp=1&amp;url=https%3A%2F%2Fupscri.be%2Fd8f5c0%2F&amp;image=https%3A%2F%2Fe.enpose.co%2F%3Fkey%3DdRXnS9Gplk%26w%3D700%26h%3D425%26url%3Dhttps%253A%252F%252Fupscri.be%252Fd8f5c0%252F%253Fenpose&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/2eb47df435a6a04de66cb3cadc7fde09/href">https://medium.com/media/2eb47df435a6a04de66cb3cadc7fde09/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b48c8c4160c7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What’s wrong with Wi-Fi?]]></title>
            <link>https://medium.com/@DomTheAssistant/whats-wrong-with-wi-fi-f29a84f3eac6?source=rss-1a84e29b81aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/f29a84f3eac6</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[transfer-rates]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[wifi]]></category>
            <category><![CDATA[home-technology]]></category>
            <dc:creator><![CDATA[Dom]]></dc:creator>
            <pubDate>Fri, 18 May 2018 11:11:03 GMT</pubDate>
            <atom:updated>2018-06-26T10:58:13.098Z</atom:updated>
            <content:encoded><![CDATA[<h4>Why is the Quality of Experience for home Wi-Fi so low?<br>Can it be improved?</h4><p><em>Written by Magnus Olden</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ypxlzpdFX18zkVZx-AaWeg.png" /></figure><p>Domos have over the last year measured Wi-Fi Quality of Experience metrics in over 100 000 homes on a second by second basis. Through this data gathering and extensive research we’ve learned to identify the cause and effects of low QoE.</p><p>The major pain is data transmitted slowly, or with low data rates as is industry jargon. We show that the <em>primary</em> cause of low data rates is <strong><em>not</em></strong> poor signal-strength as is popular belief, but end-user devices’ suboptimal selection of data rates. We show that the data rates set by devices are significantly lower than those set by the routers and that device types and device vendors show distinct behavioral patterns. And that it should be possible, though complex, to significantly increase speeds through configurations on the router.</p><p>Understanding Wi-Fi QoE is the core of Domos’ research, and we invite others to join us in attempts to optimize it on current and future hardware and protocols. Our preliminary results shows that with machine learning and statistical techniques, optimization is possible by recognizing device types. We can learn to manipulate the behavior of their rate adaptation algorithm, finding their strengths and weaknesses in different types of environments and for different types of usage.</p><p>The sections are structures as following:</p><ol><li>Evidence for low data rates</li><li>Evidence for low rates not caused primarily by low signal strength, radio noise or congestion</li><li>Shows different and suboptimal device behaviour</li><li>Discussion on optimization</li><li>References, methodology, appendix explains data rates</li></ol><h3>1. Rates are low!</h3><p>In the wild and in people’s’ homes, rates are a lot lower than they should be, both on 2.4 GHz and 5 GHz. Don’t take my word for it, let’s first review some research from Google Fiber [1]. This graph looks at number of devices connected at each speed for both 2.4 GHz and 5 GHz.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/651/0*CBvAZALHZaiMU3L1." /><figcaption>GFiber Wifi Data, Avery Pennarun (<a href="https://apenwarr.ca/diary/wifi-data-apenwarr-201602.pdf">https://apenwarr.ca/diary/wifi-data-apenwarr-201602.pdf</a>)</figcaption></figure><p>It shows that the majority of devices on 2.4 GHz are connected on less than 50 Mbps rate and less than 100 Mbps for 5 GHz. Keep in mind that these are the data rates, and is far from translating directly to actual TCP throughput. The rule of thumb says TCP throughput will be about half of what the data rates are, and that’s assuming no neighbours on the same channel and no bluetooth or other non-Wi-Fi radio devices on the same frequencies are transmitting and that the traffic is from a single device, far from the reality most people face.</p><p>From our own research, we have gone one step further, and don’t simply look at what rates devices are connected on, but how much data is sent on each rate from each device.</p><p>The y axis represents the percentage of data on each rate from the x-axis. I’ve circled the main clusters. This graph is for 5 GHz exclusively.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ya7kNIi-L0hK7EuZ." /></figure><p>We see that most traffic is sent on rates below 100 Mbps, and almost 50 % of the traffic is sent on rates lower than 50 Mbps. This is obviously far from the physical rates of 802.11ac, which is 3433 Mbps (2600 Mbps for 3 MIMO, which most of the routers in our data have got). So almost 50 % of traffic is being sent on rates that are less than 0.15 % of the max rate, this is a lot worse than the industry seems to realize. The actual QoE for an end user will actually be a lot worse than it looks by these graphs. The amount of airtime that is being spent on data on very low rates are staggeringly high (and is a problem worthy of another paper). We are left with a single question.</p><h3>2. Why?</h3><p>Why are the rates so low? Signal strength? Noise? Congestion? Hidden Node problems? The general idea is that loss in signal strength should lead to data being sent on lower rates, as low rates have greater resilience for poor signal and noise. Does theory hold up to practice?</p><p>Lets review another graph from Google Fiber Wi-Fi project:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/810/0*-DhTc-PvZ9ckn35A." /><figcaption>GFiber Wifi Data, Avery Pennarun (<a href="https://apenwarr.ca/diary/wifi-data-apenwarr-201602.pdf">https://apenwarr.ca/diary/wifi-data-apenwarr-201602.pdf</a>)</figcaption></figure><p>This graph shows the rates that thousand of devices are utilizing against their signal strength (rssi). While there is a correlation between rssi and rate, it is surprisingly low, and the spread of rates at nearly every level of signal strength is huge. There are plenty of traffic on low rates even with high RSSI both on 2.4 GHz and 5 GHz. Again, reaffirming that most traffic is sent on low rates and showing that signal strength alone is far from explaining low rates.</p><p>This is also supported by our own research, we do however have cooler 3d graphs. Here a graph where 2.4 GHz and 5 GHz combined. The X axis represents rates (lowest left, highest right), the Z axis is SNR (1 lowest in front, 50 highest in the back), Y axis, the amount of traffic.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*lEUC27g66aDJTxoQ." /></figure><p>Again we can observe that here is a correlation between SNR and rate, but it is low and we see next to no traffic on the highest rates.. The five largest peaks of data are from rates of 78, 39, 24, 6.5, 1 Mbps, and with the exception of the highest 78 Mbps, they have a high variety of SNRs.</p><p>What have we learned thus far? With the exception of the highest rates not being used when the signal strength is low, which is expected behaviour, there is still a lot of unexplainable behavior. Why are there so much data on low rates with high SNR?</p><p>Next on the list for expected culprits is congestion, other routers or other users on the same frequency causing packet collision, packet losses and faulty packets. We have done the same measurements against congestion. Y axis here is % time where airtime was used by others.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*SRPk7sXaDUyafPD_." /></figure><p>We see a close to zero correlation between congestion level and the Data Rate. In total SNR and Congestion can explain far less than 50 % of the low rates. Then what?</p><h3>3. The Devices</h3><p>A big piece of the puzzle is the observation that there is massive difference between the rates that traffic is sent on;</p><p>From the devices to the router. And,</p><p>From the router to the devices</p><p>The traffic from the device to the router is sent on rates a lot lower than the rates from the router to the device. Splitting the previous graph into RX (left), from device to router and TX (right), from router to device is shown in these two graphs. The graphs follow the same format as the first.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*k2a_U7QKHCi2lLqI." /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*j-iZdKYk4tvEkupC." /></figure><p>From devices to the router (left graph) almost all the traffic, actually more than 70 %, is sent on 21 Mbps or less. From the router to the devices (right graph) more than 70 % of the traffic is sent on 39 Mbps or higher. On average, the traffic from the router, is sent on more than four times the speed than the other way.</p><p>Observing this we are left with the questions.</p><ul><li>Why are devices sending data on low rates for no apparent reason?</li><li>Why is there such a great difference between from and to the router?</li><li>What can be done about this from the router? Can the devices be forced to send traffic on higher rates? Will upgrading the router help?</li></ul><p>It is easy to dismiss the upload traffic, as it is far less than download. However, remember upload and download shares the same airtime and …</p><p>For a brief explanation of what low rates are and how rates are set. Please see appendix .</p><p>Other observations from the wild:</p><ul><li>Devices tend to lower their rate from very high to very low.</li><li>Traffic sent near the PHY max rates are as common as unicorns.</li><li>Some devices seem to simply stick to low rates no matter what.</li><li>Different devices handle different environments differently. Some are very susceptible to noise, some to signal and some to congestion.</li></ul><p>Fully explaining all of this is a difficult and comprehensive task. Which is what we are currently researching. The next section is about some distinct device behaviour we are observing.</p><p><strong>Examples of device behaviour</strong></p><p>So how differently are devices behaving? Let’s do a quick review of Apple devices Vs. every other device (about 45 % vs 55 % of the traffic) in a household. The next graph shows Apple devices only. What is interesting to observe is that nearly all traffic is sent on the same rate 24 Mbps. They do this with SNRs ranging from 10–48.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*S1ckN2S8txaZqE9J." /></figure><p>On the other hand, the rest of the devices choose higher and lower rates. As we can see in the next graph, they use both higher rates, 79 Mbps, and lower, with a lot of traffic on 1, 6.5 and 21 Mbps.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*t8_wWucQ_g-pNgCG." /></figure><p>Which is better? The Apple way or the highway? While the average rate is about the same, the airtime consumed by devices sending on the lowest rates is extreme! Incidentally 24 Mbps is about what is needed to facetime in HD, with a bit of leeway.</p><p>Chromecast also shows a very distinct behaviour. As seen in the next graph which shows upload from chromecasts exclusively we can see that the upload is almost only on the lowest possible rate (1 Mbps for 2.4 GHz). This can be observed in the next graph:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*A6dysGrba590eett." /></figure><p>Chromecasts sends a small amount of data at regular intervals (every 20th second). On the other hand, the next graph shows download to chromecasts, and it is sent on relatively high rates.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*qepaQaC762m8JuPS." /></figure><p>Chromecasts don’t upload a lot of data, but it is worth noting that even though the chromecasts downloaded 100 times more data than they uploaded. They roughly consumed the same amount of airtime uploading and downloading.</p><p>These results indicates that there are greater differences between devices than most research accounts for, and certainly worth looking more into.</p><h3>4. How to optimize</h3><p>Certain routers can force devices to choose a set of rates, or some rates can be blocked. This allows for optimization and improvement. This is a very, very complex problem. To even begin to optimize rates you need to identify:</p><ol><li>The device</li><li>The Rate Adoption Scheme</li><li>Strength and weaknesses of different Rate Adoption Schemes in different environments</li><li>How the Rate Adoption Scheme reacts to having certain rates blocked</li></ol><p>These insights needs to be combined with live data regarding;</p><p>SNR, congestion, connected devices, packet loss, retransmit, etc.</p><p>Only when all of these factors are know, can you properly optimize. We are now in the process of researching and deploying a Reinforcement Learning algorithm in a test environment allowing it to set rates for different device types and learn from the results. We are also working with the data with more classical statistical approaches and other machine learning algorithms.</p><p>There are also other configurations that can be set that improves the data rates of the devices. For example Short Guard Intervals, MIMO and Frame Aggregation configurations varies in performance for different devices and different environments.</p><p>Preliminary testing shows that it is possible to manipulate and improve devices rate selection through router-settings.</p><p>We are currently researching and trying to understand and optimize this, and are looking for research partners.</p><h3>Methodology</h3><p>The rate of the last package is sampled every second. We have verified the results with Wireshark sniffing and found that, even though rate is highly variable, the sampling frequency captures the relevant trends.</p><p>The SNR/RSSI is sampled every 5th second</p><p>Congestion levels are sampled every 5th second.</p><p>For more details, please ask.</p><h3>References</h3><p>[1] GFiber Wifi Data, Avery Pennarun</p><p><a href="https://apenwarr.ca/diary/wifi-data-apenwarr-201602.pdf">https://apenwarr.ca/diary/wifi-data-apenwarr-201602.pdf</a></p><h3>Appendix</h3><p>Why are rates reduced at all? What it all comes down to is the nature of digital transfer, the available information, memory and storage constraints in the network chips and how different networking chips choose their rate.</p><p>What are the different rates? Disregarding MIMO and simplifying a bit, a higher data rate means packing more data into a single radiowave, making it possible to send more bits per radio wave (this is done by increasing the resolution of the amplitude and phase of the radio wave). The downside is that the signal becomes more susceptible to noise and degradation.</p><p>Every package sent has a checksum, making it possible to calculate whether or not every bit in the package is correct. If an error is found, the network chip ask for a retransmit. If a single or a series of packages is lost (no acknowledgement is returned) or has faulty bits, the networking chip chooses a lower rate, giving it a higher chance of succeeding.</p><p>How the next rate is set is known as the rate adaptation scheme. There are two main ways of implementing these. Those rate adaption schemes that differentiate between losses from degraded signal and collisions and those who don’t. Exactly how this is implemented is all up to the networking chip. They are all closed sourced so we have to .</p><p>What we know is that all network chips have:</p><ol><li>Little historical information</li><li>Little information about the environment</li><li>Is likely optimized for its own performance and in simulations or lab environments</li><li>There are almost always trade-offs. Throughput vs Reliability vs Memory/Storage/CPU usage and price</li><li>802.11 Limits what data is, and can be shared between the router and the devices. Limiting possible solutions</li></ol><p>To summarize, higher rates are more error prone, and there are many constraints preventing devices from setting a perfect rate. Especially increasing the rate after its been decreased seems to be a very slow progress for most devices. We are observing some devices that needs a reconnect to start sending data at high rates. The result of this is that a lot of traffic is sent on sub-optimal rates.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2Fd8f5c0%3Fas_embed%3Dtrue&amp;dntp=1&amp;url=https%3A%2F%2Fupscri.be%2Fd8f5c0%2F&amp;image=https%3A%2F%2Fe.enpose.co%2F%3Fkey%3DdRXnS9Gplk%26w%3D700%26h%3D425%26url%3Dhttps%253A%252F%252Fupscri.be%252Fd8f5c0%252F%253Fenpose&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/2eb47df435a6a04de66cb3cadc7fde09/href">https://medium.com/media/2eb47df435a6a04de66cb3cadc7fde09/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f29a84f3eac6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Customer Service — From Data Buffet to Intelligent Assistance]]></title>
            <link>https://medium.com/@DomTheAssistant/customer-service-from-data-buffet-to-intelligent-assistance-296fc819dfb0?source=rss-1a84e29b81aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/296fc819dfb0</guid>
            <category><![CDATA[intelligent-assistant]]></category>
            <category><![CDATA[internet]]></category>
            <category><![CDATA[wifi]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[customer-service]]></category>
            <dc:creator><![CDATA[Dom]]></dc:creator>
            <pubDate>Thu, 25 Jan 2018 14:26:49 GMT</pubDate>
            <atom:updated>2018-06-05T14:23:30.297Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DF5ynWqU1CQcR9MSH4wJNg.png" /></figure><p>Our first touchpoint with a new operator tends to be their customer call center. They are the ones feeling and sharing the customer pain — but are way too often left feeling frustrated and feeble by the lack of actionable insight in their traditional support tools.</p><p>We began testing with our first operator customer a bit over a year ago. It is interesting to look back at how our approach and UX has evolved over this time — partly as we have built more intelligence into the system but even more so as we have had the opportunity to work in close dialogue with the customer care teams, monitoring their workflow and being involved in their strategic improvement process.</p><p>This has been a product journey starting from Data Visualization and ending (so far) with Intelligent Assistance. It is also a journey of radically simplifying UX for customer care reps, helping management achieve a strategic goal of dramatically reduced resolution time.</p><h3>1st Gen: The Full Buffet</h3><p>In our first version of Diagnostics, we did many of the typical mistakes we see with our competitors. We were so proud of all the data we collected so we wanted to show them all — at once. This clearly had some value as we were able to provide data on Wi-Fi environment that was until then unknown in the industry. Unfortunately the typical customer rep with an angry customer on the line was generally overwhelmed, requiring extensive training beforehand and long processing time with customer on hold.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*bOM8J7b0sw1AkCwo." /><figcaption>Because who does not know about Airtime?</figcaption></figure><p>We added a color-coded logo on each graph to indicate whether there was reason to investigate further. But the deeper investigation was still mainly visualization of data:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/967/0*oCQFk9cUOq7yyhV8." /><figcaption>Current status — Quite obvious what to fix right?</figcaption></figure><p>To top it off, we presented this elegant little device list. Honestly, I have no idea what we were thinking:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*sNm8RNdE42FnGOI9." /><figcaption>Data unleashed — Never again…</figcaption></figure><h3>2nd Gen: From Data Dump to Actionable Insights</h3><p>While our 1st generation Diagnostics was well received by the experts and techies, it was also evident that it was far too complicated for the generalists. We noticed the average 1st line rep regularly missed the clear indications that they could have read out of the data if they had the necessary processing time and training. The obvious solution was to reuse the features we had built into Dom, the intelligent assistant that had been developed in parallel for the consumer self-service app.</p><p>So in this version the center part of the screen highlight the actual issues experienced by this customer, and suggestions on how it could be fixed. We decided to list all possible issues and rank them to the relative impact. We also kept the top level status information, the recent events and the device list.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1010/0*SPYMbqCBKfvTjD8_." /><figcaption>Less clutter, but still a lot to interpret</figcaption></figure><p>We also incorporated the comments from earlier workshops with customer service teams. This generation included two different modes, Easy Mode and God Mode (can’t be serious all the time). Image above shows the easy mode, while God Mode included more advanced modules, graphs and actionable settings. God Mode was everything 2nd line wanted, but Easy Mode was still not easy enough. 1st line reps ended up not using it as intended, partly because there was no clear answer to what the issues were and partly they simply did not feel confident on how to use and understand the view. So we went back to the drawing board….</p><h3>3rd Gen: The Intelligent Assistant</h3><p>As the 2nd generation was a huge improvement, we saw through our observations that it still didn’t quite fit the workflow structure in the typical call center operation. Their operational goal of 1st line of support was essentially just to figure out nature of call. They typically followed a set script and went through a set workflow. Overall goal is to significantly reduce handling time, in which the issue was either resolved or passed on to expert level support.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*yALuOMyJbJjGb01j." /><figcaption>Guided entry to router diagnostic</figcaption></figure><p>Instead of focusing on our data and insight, we now focus on the workflow and personal communication with the customer. The entry point is now a simple, 3 color option indicating the typical information that customer share immediately in the call — is this a persistent or a new issue, and has this been a recurring inquiry. We use this input to guide the diagnostics, and come up with one clear and actionable response like in the example below. If no clear conclusion can be reached, we recommend upgrading the customer issue to 2nd line of support. During this transition we of course keep the context and input from 1st line to ensure efficient handover.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7P4pn6SyX8FEJagCFLEnfg.png" /><figcaption>Guiding CS from identifying the issue to how they can explain it and solve it for the customer</figcaption></figure><p>We wanted Dom to be as central to this solution as to the end-users self-service app. In the end, both interfaces served the same purpose — guide users to locate current issue and how to solve it. By focusing the interface on the most probable issue, the need for training, interpreting and reviewing information was no longer needed. Just choose the customer situation and you’re ready to assist.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2Fd8f5c0%3Fas_embed%3Dtrue&amp;dntp=1&amp;url=https%3A%2F%2Fupscri.be%2Fd8f5c0%2F&amp;image=https%3A%2F%2Fe.enpose.co%2F%3Fkey%3DdRXnS9Gplk%26w%3D700%26h%3D425%26url%3Dhttps%253A%252F%252Fupscri.be%252Fd8f5c0%252F%253Fenpose&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/2eb47df435a6a04de66cb3cadc7fde09/href">https://medium.com/media/2eb47df435a6a04de66cb3cadc7fde09/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=296fc819dfb0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Optimizing Wi-Fi in congested environments using AI]]></title>
            <link>https://medium.com/@DomTheAssistant/optimizing-wi-fi-in-congested-environments-using-ai-a61721530392?source=rss-1a84e29b81aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/a61721530392</guid>
            <category><![CDATA[mesh-networks]]></category>
            <category><![CDATA[wifi]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[analytics]]></category>
            <dc:creator><![CDATA[Dom]]></dc:creator>
            <pubDate>Fri, 27 Oct 2017 12:06:39 GMT</pubDate>
            <atom:updated>2018-06-05T14:23:51.662Z</atom:updated>
            <content:encoded><![CDATA[<h4>Artificial Intelligence increases Wi-Fi throughput by 100% in typical urban apartment complex</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MbK7iyRB6zzQ-vhY8w2krg.png" /><figcaption>How do you solve Wi-Fi for an area where everyone conflict and disrupt each other?</figcaption></figure><p>Over the summer we deployed and refined our <a href="https://medium.com/domos-creating-the-home-it-assistant/area-optimized-wi-fi-and-how-to-do-it-13a09d8658db">“dense area” algorithm</a> in a typical apartment building in Oslo. The ISP had received serious complaints from their customers in the building after upgrading internet connections to 100/100 fiber. We controlled about 50% of the access points, all single-band (i.e. 2.4GHz only).</p><p>Our “dense area” algorithm focus solely on reducing congestion and interference on the 802.11 bands. Hence, we wanted to measure the total throughput over Wi-Fi in the building as well as the total airtime consumed. As long as the transfer rates are constant and demand is strong, increase in airtime availability will translate directly to better Wi-Fi capacity.</p><p>The results below show total traffic and airtime by hour, measured over two weeks before and after deploying the algorithm:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_6Q40Y6mPqGHmkuIm4PDrQ.png" /><figcaption>Improvement in total hourly traffic</figcaption></figure><p>Conclusion is that in high demand periods our algorithm increased both airtime availability and Wi-Fi capacity by roughly 100%.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7eX2LbQHS3KySCUj6qrbGg.png" /><figcaption>Improvement in hourly airtime usage</figcaption></figure><p>So how did we achieve this? By deploying two distinct optimization algorithms:</p><ol><li>Optimizing channel settings according to our genetic algorithm approach as described in <a href="https://medium.com/domos-creating-the-home-it-assistant/darwin-the-sysadmin-3c89cabb21e6">earlier blog post</a></li><li>By teaching the access points to use their “indoor voice” (a method we have also <a href="https://medium.com/domos-creating-the-home-it-assistant/area-optimized-wi-fi-and-how-to-do-it-13a09d8658db">outlined</a> earlier) by reducing transmit power without reducing signal quality or transfer rates.</li></ol><p>As a result, we were able to reduce the average number of AP channel conflicts from 7 to 3, as illustrated in the histogram below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bi3V3utLII7zX9hfirFD-g.png" /><figcaption>Reduction in competing AP channels</figcaption></figure><p>The transmit power settings were reduced dynamically to an optimal state as represented in the histogram below (100 mW is default and maximum allowed):</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L5S9zBzOkQ8V-Bf0pWaw1Q.png" /><figcaption>Less than half of the routers needed max power settings</figcaption></figure><p>This algorithm is now available as an end-to-end automated service where it will:</p><p>1) <strong>Detect</strong> dense areas based on routine scans of the environment</p><p>2) <strong>Evaluate</strong> the improvement potential based on congestion level and proportion of access points under our control</p><p>3) <strong>Optimize</strong> throughput continuously as new access points are added or there are other changes to the environment.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2Fd8f5c0%3Fas_embed%3Dtrue&amp;dntp=1&amp;url=https%3A%2F%2Fupscri.be%2Fd8f5c0%2F&amp;image=https%3A%2F%2Fe.enpose.co%2F%3Fkey%3DdRXnS9Gplk%26w%3D700%26h%3D425%26url%3Dhttps%253A%252F%252Fupscri.be%252Fd8f5c0%252F%253Fenpose&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/2eb47df435a6a04de66cb3cadc7fde09/href">https://medium.com/media/2eb47df435a6a04de66cb3cadc7fde09/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a61721530392" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>