The performance goals for 5G are ambitious, impressive, and will be world-changing. Achieving them is also a significant challenge, and as the world races to develop, test, and deploy 5G technology, existing approaches using traditional design methodologies are starting to show some of their limits.

Need for a new Approach

Customer thirst for ever-increasing mobile data is driving the urgency to move from 4G to 5G. While 5G has the potential to offer a significantly enhanced user experience, its core wireless technology heritage is over 20 years old. Moore’s Law has reduced the rack scale systems of the 1990s to our present-day smart phones, but the core algorithms have not evolved. The result is 5G systems that consume dramatically more power than desired, achieve lower data rates than planned, and leave predictable user and channel estimation effects untapped. DeepSig addresses these problems by replacing specific traditional wireless algorithms with deep learning AI to dramatically reduce power consumption and improve performance. This is a fundamentally different and far more performant approach than other AI-based approaches that focus primarily on network management and scheduling. DeepSig has productised this approach with its OmniPHY™ software and is working with partners to provide unprecedented levels of performance in current and next generation RAN systems.

OmniPHY’s Technology Heritage

Our OmniPHY software leverages key advances made in machine learning over the past few years in the fields of computer vision, natural language processing, and voice recognition but fundamentally applies these concepts to key baseband processing tasks. We leverage software and ideas pioneer by companies including Google and Facebook which led to drastic improvements in computer vision and apply these likewise to key hard to model problems in wireless which degrade performance in current systems. The key insight is that insufficient wireless models and simplified human engineering approaches can often be replaced by deep (i.e., many-layered) neural networks that learn directly from real-world data and continue to update in the field. Instead of using engineering approximations and simplifications, a deep neural network embraces complexity and learns a near optimal solution for key performance metrics such as data throughput, multi-user capacity, latency minimization and power consumption reduction. This approach has the added benefit of increasing the overall quality of service, resilience to lower cost or imperfect hardware components, and which reduces things like dropped calls and link degradation in poor weather.

Machine Learning And Signal Processing

5G employs industry standard RAN waveforms, such as the “5G New Radio Air Interface”, which are developed by the 3GPP International Committee - so OmniPHY can’t develop entirely new waveforms for 5G (note that this is not the case for 6G) and must comply with the specification in order to ensure interoperability.  Thus, to achieve performance improvements in 5G, OmniPHY substitutes AI for existing processing algorithms while maintaining compatibility with the 3GPP 5G standard.

To illustrate how this works, we'll use Massive MIMO as an example, where the smaller antenna systems of 4G are replaced by arrays of up to 192 antennas employing either fully digital MIMO processing or combined analog/digital sub-array processing. The processing required to fully utilize Massive MIMO in both of these cases has turned out to be tremendously expensive in terms of computation, which also means it is tremendously power hungry. In fact, when operators tested this technology they discovered that their battery backups went from days worth of capacity to just a few hours in some cases. Using OmniPHY, DeepSig is able to replace key high-complexity processing steps with efficient AI in the form of compact deep learning neural network inference engines that achieve better fidelity and functionality with substantially less power consumption.

This process, replacing simplified analytic algorithms with AI-based equivalents that use machine learning to adapt and improve after deployment will bring significant computational and power benefits to many areas of the 5G stack, allow for cost reduction and relaxation of hardware linearity requirements, and will help to reduce cost by increasing cell and user capacity on current hardware and spectrum resources.

Enhancing the 5G-NR Ran with Machine Learning

DeepSig’s enhanced algorithms within the 5G RAN help reduce power, reduce component cost, increase device density and performance, and make operation and deployment of 5G BTS deployments cheaper and more autonomous in terms of both OpEx and CapEx.

DeepSig is investing heavily in the application of AI for 5G and other consumer wireless technologies, and is rapidly developing, vetting and productising these capabilities with our OmniPHY software product tightly coupled within NR L1 RAN implementation. Real-time demonstrations of our solutions running over the air on NR test equipment within a partial L1 stack are now available in wideband NR configurations. If you would like to learn more, please get in touch using the contact form below!

Inquiries

DeepSig is investing heavily in the application of AI for 5G and other consumer wireless technologies, and is productizing these capabilities with our OmniPHY software product. If you would like to learn more, please get in touch using the contact form below!