Using Data Science to Optimize the Wireless Experience and Minimize WLAN Costs
The explosive growth of smart mobile devices, applications, and Internet of Things (IoT) has created a big challenge for legacy wireless LAN products, which were all designed when smartphones and tablets didn’t exist and cloud platforms like AWS were still in their infancy.
For example, smart devices have substantially increased the number of users on the wireless network. This not only creates capacity, coverage, and interference issues that adversely impact performance, but it severely complicates Wi-Fi troubleshooting. There are way more mobile hardware platforms, operating systems, and applications today than a decade ago, which cause IT to constantly react to user issues vs. proactively planning ahead of them. Using packet sniffers and manual processes to identify and remediate problems across all these different variables just won’t scale, creating the need for a better solution for wireless operations in today’s smart device era.
In addition, because mobile devices have become the predominate compute platform, Wi-Fi networks have moved from nice-to-have to business-critical in most environments. Flaky wireless coverage is no longer acceptable, nor is an inconsistent experience across users and devices. When problems happen, IT needs to rapidly respond and resolve them.
To achieve business criticality, the following age-old Wi-Fi operational challenges need to be addressed once and for all:
- Packet sniffing is expensive, time consuming, and often ineffective. Many wireless problems are ephemeral, disappearing shortly after they arise based on changing user and environmental conditions. Sending techs onsite to reproduce the problem can be expensive, and often yields lackluster results as the data needed to reproduce and resolve an issue can be long gone.
- It is difficult to pinpoint the root cause of problems. Often times when a mobile user cannot connect or is receiving sub-par performance, the wireless network is the first thing that is blamed. However, it could just as easily be an issue with DNS, DHCP, authentication servers, or a variety of other things. Administrators need a quick and easy way to identify root causes for fast remediation.
- Administrators lack visibility into what users are actually experiencing. Traditional WLAN systems take a network-centric view of the world. While they are good at telling you what an Access Point is experiencing, they provide little insight into the wireless experience from the users’ perspectives. This complicates troubleshooting and makes it all but impossible to monitor and enforce service level expectations for key metrics like connect time, capacity, coverage, and roaming.
Thanks to machine learning and other AI technologies, it is finally possible to address these issues in a seamless and scalable manner. Isolate the root cause of problems more quickly. Predict problems before they occur. Set, monitor and enforce user service levels. This is all finally possible with AI-driven WLANs, delivering a better wireless experience for IT administrators, which translates to an amazing Wi-Fi experience for mobile users.
Key Requirements for AI-driven Wireless
There are four key components to building an AI engine for a WLAN: data, structure and classify, data science and insight. Let’s take a closer look at each.
Data: Just like wine is only as good as the grapes, the AI engine is only as good as the data gathered from the network, applications, devices and users. To build a great AI platform, you need high-quality data — and a lot of it.
To address this well, one needs to design purpose-built access points that collect pre- and post-connection states from every wireless device. They need to collect both synchronous and asynchronous data. Synchronous data is the typical data you see from other systems, such as network status. Asynchronous data is also critical, as it gives the user state information needed to create user service levels and detect anomalies at the edge.
This information, or metadata, is sent to the cloud, where the AI engine can then structure and classify this data.
AI primitives: Next, the AI engine needs to structure the metadata received from the network elements in a set of AI primitives.
The AI engine must be programmed with wireless network domain knowledge so that the structured metadata can then be classified properly for analysis by the data science toolbox and ultimately deliver insights into the network.
Various AI primitives, structured as metrics and classifiers, are used to track the end-to-end user experience for key areas like time to connect, throughput, coverage, capacity and roaming. By tracking when these elements succeed, fail or start to trend in a direction, and determining the reason why, the AI engine can give the visibility needed to set, monitor and enforce service levels.
Data science: Once the data has been collected, measured and classified, the data science can then be applied. This is where the fun begins.
There are a variety of techniques that can be used, including supervised and unsupervised machine learning, data mining, deep learning and mutual information. They are used to perform functions like baselining, anomaly detection, event correlation and predictive recommendations.
For example, time-series data is baselined and used to detect anomalies, which is then combined with event correlation to rapidly determine the root cause of wireless, wired and device issues. By combining these techniques together, network administrators can lower the mean-time-to-repair issues, which saves time and money and maximizes end-user satisfaction.
Mutual information is also applied to Wi-Fi service levels to predict network success. More specifically, unstructured data is taken from the wireless edge and converted into domain-specific metrics, such as time to connect, throughput and roaming. Mutual information is applied to the service-level enforcement metrics to determine which network features are the most likely to cause success or failure as well as the scope of impact.
In addition, unsupervised machine learning can be used for highly accurate indoor location. For received signal strength indicator-based location systems, there is a model needed that maps RSSI to distance, often referred to as the RF path loss model. Typically, this model is learned by manually collecting data known as fingerprinting. But with AI, path loss can be calculated in real time using machine learning by taking RSSI data from directional BLE antenna arrays. The result is highly accurate location data that doesn’t require manual calibration or extensive site surveys.
AI-driven virtual assistants: The final component of the AI engine is a virtual assistant that delivers insights to the IT administrator as well as feeds that insight back into the network itself to automate the correction of issues, ultimately becoming a “self-healing network.”
The use of a natural language processor is critical to simplify the process for administrators to extract insights from the AI engine without needing to hunt through dashboards or common language interpreter commands as legacy systems that lack AI do. This can drive up the productivity of IT teams while delivering a better user experience for employees and customers.
Wi-Fi with Assurance
By incorporating AI and data science with wireless expertise, the following benefits are achieved:
No more manual packet sniffing
When a user is experiencing a network anomaly, the WLAN system can automatically detect it and start capturing packets, a concept known as Dynamic PCAP (dPCAP). This enables you to rewind back in time to see what was going on in the Wi-Fi network and the mobile device when the anomaly was detected.
No more sending techs onsite with sniffers to chase problems that might not even exist anymore. The data needed to fix the problem is always available and at your fingertips, which reduces your IT costs and minimizes the Mean Time to Repair (MTTR) wireless problems.
Easy and Accurate root cause analysis
With machine learning technology in place, the WLAN can dynamically collect information from all Wi-Fi mobile devices and correlate events for quick root cause identification.
By looking at each and every mobile user’s RF packets from inside the cloud, you can easily identify if the user is having a “connection”, “coverage”, “capacity”, “throughput” or ”roaming” issue. Or, you can identify if it is not a wireless issue at all – i.e. maybe it is a DNS, DHCP, WAN or authentication problem.
Visibility into user service levels
AI-driven WLANs can collect real-time state information from every mobile device on the network. This has numerous benefits, among them is the ability to set service level expectation (SLE) thresholds for all the major things that impact wireless performance, which include:
- Connection
- Coverage
- Capacity
- Roaming,
- Throughput
- Latency
- Jitter
If any of these parameters are violated, you are proactively given insight into the reasons why, top mobile devices affected, and top wireless networks affected.
Data science meets wireless wizardry
The time has come for wireless operations to move from a reactionary troubleshooting mode to a proactive one where actions can be taken that avoid problems before they arise. Plus, intelligence must be built into the wireless network so automated changes can be made in real-time, continuously optimizing Wi-Fi performance for every individual user and environment.
Wireless networks are more business-critical than ever, yet troubleshooting them is more difficult every day due to an increasing number of different devices, operating systems and applications. AI-driven WLANs are the answer, enabling businesses to keep up with the soaring numbers of new devices, things and apps in today’s connected world.
