Skip to main content


Machine Learning in the Physical (PHY) Layer

Improving 5G Open vRAN with improved efficiency and higher performance

As 5G base stations move from proprietary hardware to virtualization on commodity servers, OmniPHY-5G brings the advantages of artificial intelligence and machine learning to the 5G physical (PHY) layer. The PHY layer transforms wireless signals to data bits and neural networks improve user data rates while reducing base station power consumption. 

Delivering drop-in neural network replacements for Intel® FlexRAN™ software upper PHY functions, we provide state-of-the-art power and spectral efficiency to the 5G experience, readily adaptable to any virtualized 5G base station software suite.


MIMO and Uplink Reception

2X Reduced Compute Load

OmniPHY-5G provides an optimized neural network leveraging existing Intel® Xeon™ processors with Deep Learning Boost (DLBoost) on  Intel® FlexRAN™ platforms.  DeepSig’s replacements for uplink (PUSCH) receiver processing in the upper PHY reduces compute load by up to 2X allowing for reduced carbon footprint and more sectors per server.

Improved Spectral Efficiency – Up to 2x Throughput Increase

OmniPHY-5G enables additional MIMO capacity and user count to boost performance and capacity in FlexRAN™ distributed unit (DU) servers, allowing for higher throughput, enhanced coverage, and improved user experience.

Direct neural receiver deployment into DU

Seamless drop-in integration allows Intel® FlexRAN™ users to get going with OmniPHY-5G overnight, and immediately maximize the PHY performance. Replacement of traditional signal processing techniques and no code modifications required to leverage the advantage of pre-trained neural networks.

Massive MIMO SRS Receiver

4x Accuracy Improvements of mMIMO Channel estimates

Enable more accurate beam weights and improved multi-user capacity with enhanced throughput per user.

Efficiently schedule and steer beams to an increased user count per sector, simultaneously.

When using Massive MIMO, 5G uses beamforming to simultaneously deliver additional bandwidth to multiple users to increase capacity. Sounding reference signals (SRS) are transmitted from phones to ensure these beams are scheduled and formed accurately, impacting both the resulting capacity and computational load on the DU server.

5x Reduction in SRS Processing Compute Load

Receive more affordable and scalable deployment of mMIMO with DeepSig’s OmniPHY-5G.

What’s to Come?

Continuous Monitoring and Improvement of Neural Receiver Performance

Continually optimize OmniPHY-5G performance on real world data and monitor impact and benefits of neural processing in the L1. Run OmniPHY-5G in a RIC xApp or directly on the DU server.

Enhanced Beam Weights and Scheduling

OmniPHY-5G will further improve mMIMO performance and efficiency through data-driven ML enhancements. Beam-weight computation and multi-user scheduling approaches are currently under test and development.

Related Products

Discover more with DeepSig


© 2024 DeepSig Inc. All rights reserved.