Harnessing the Power of AI-driven Spectrum Sensing for 5G
Peter Carson
Product Marketing | Thought Leader | Team Builder @ Lightmatter ? AI ? Compute ? Connectivity ? Cloud
Spectrum Sensing is an important capability for understanding what kind of wireless radio activity is present in frequency bands. This knowledge, or Spectrum Awareness, enables and enhances a range of applications, from spectrum sharing and mapping to fault detection, security and regulatory monitoring. AI/ML-based signal detection and classification significantly advances the speed, accuracy, sensitivity and ease of building and deploying a wide range of sensing solutions at the edge for a diverse set of use cases.
Signal Detection and Classification?
In wireless systems like Cellular 4G LTE, 5G-NR, Wi-Fi, and Bluetooth, the frequencies, bandwidths and timeslots are pre-coordinated, meaning that both the transmitter and receiver know which bands have been allocated to them and where to find the signal in the spectrum. This allows the receiver to focus on specific frequency bands and specific protocols (e.g., preambles) for signal detection. When unwanted signals or interferers impair 5G reception, they often do not follow the same predictable patterns and can be very difficult to pinpoint, occurring randomly in time and frequency. Typical network-based monitoring relies primarily on detecting degradation in key performance indicators (KPIs) such as throughput or received signal quality degradation but often point to a vague problem and unexplained cause. Dispatching technicians for an on-site diagnosis with test and measurement equipment can be costly and unreliable, particularly for intermittent emissions that may not occur predictably or during a site visit.
In federal or industrial networks, where securing communications channels and preventing unwanted surveillance are paramount, radio spectrum awareness becomes even more vital. Certain events may indicate attacks such as wireless spoofing, jamming, or injection, where malicious entities attempt to manipulate or disrupt the wireless system. In many systems, such a signal may be virtually indistinguishable from a legitimate one and may require instantaneous detection, classification and mitigation, leaving no time for human-in-the-loop analysis or reaction time. Similarly, protection from unwanted surveillance by drones or other nearby communications systems requires the ability to reliably detect and identify potentially threatening communications emissions, rapidly recognizing or localizing nearby potential threats. Such mission-critical use cases have led to rapid uptake of AI-driven wideband spectrum sensing and alerting in recent years. Such AI-driven systems can quickly learn to identify a wide range of emitters and efficiently detect such emissions.
Global Spectrum Challenge?
Radio frequency (RF) spectrum is a finite, valuable and often inefficiently used natural resource. The economics of ubiquitous mobile networks have been built on the radio propagation characteristics of the sub 6 GHz bands. Mobile network operators (MNOs) have grown accustomed to having exclusive access to large swaths of bandwidth in this sweet spot of the electromagnetic spectrum. As frequencies become congested with traffic, capacity is added through air interface upgrades and cell site and antenna densification (small cells and massive MIMO). As these approaches start to reach economically diminishing returns, the search for more radio spectrum ensues. A looming shortage of commercially available low and mid-band frequencies poses long-term challenges to the economic health of the industry. Even as legacy bands in lower frequencies are re-farmed from 4G to support 5G, they tend to be significantly narrower in bandwidth than newer mid-bands and offer limited capacity relief.
There is growing consensus among standards, regulatory and technology experts that the industry’s long-term spectrum needs cannot be met without significant advances in spectrum sharing. As an example, the global quest for a new commercial mobile spectrum has landed on bands that are used by national defense organizations around the world, most notably, 3.3-3.45 GHz, 4.4-4.8 GHz and 7.125-8.400 GHz. This has kicked off a number of spectrum-sharing studies, including those recommended in a report from The World Radiocommunication Conference 2023 (WRC-23), published by the International Telecommunication Union - Radiocommunications Sector (ITU-R). This spectrum search represents a global challenge for both commercial and government stakeholders.
The U.S. Spectrum Sharing Experiment
In the Citizens Broadband Radio Service (CBRS) band, where 4G/5G wireless networks coexist with incumbent naval radar systems, a very specialized Environmental Sensing Capability (ESC) has been implemented to detect the presence of high-power naval radar signals (the federal primary user) and report them to a cloud-based Spectrum Access System (SAS). When a radar signal is detected, the SAS administrator directs CBRS devices to different frequencies and/or to reduce their transmit power. Uncertainties about CBRS coexistence in dense coastal exclusion regions has relegated the band’s use to small scale private 4G and 5G deployments. The documented shortcomings of its sharing mechanism have led to a new CBRS 2.0 effort, expanding density and utility of the band through improved propagation model accuracy and rules. This is a step forward but what the industry needs now is a leap forward.
The dedicated ESC approach may be sufficient for the CBRS band because naval radar is generally restricted to large coastal areas and operate at high power. As a result, only a limited number of ESC sites is required to detect the presence of a single, well-defined incumbent user signal. The problem becomes more complicated in the prospective 3.1-3.45 GHz band, where 5G systems would need to coexist with critical Department of Defense (DoD) systems that operate at lower power levels over a much wider range of possible locations on the ground, in the air, at sea, and in space. A recent DoD study has recently concluded that sharing this 350 MHz of prime 3 GHz spectrum would not be feasible unless dynamic spectrum sharing (DSS) can be proven out at scale. It’s no wonder that the associated call to action for public and private sector collaboration in the White House’s?2023 National Spectrum Strategy paper, was characterized as a “moonshot” effort. This implies a much more powerful, distributed and sensitive spectrum awareness solution deployed broadly alongside radio infrastructure and capable of identifying a wide range of waveforms and protocols.
ORAN Enabled Spectrum Sensing
Let’s take a fresh look at the problem by considering an Open RAN (ORAN) enabled implementation of distributed spectrum sensing in a 5G network. O-RAN Alliance standards provide an architecture for implementation at scale by ensuring that applications built on top of the RAN, such as spectrum sensing, can leverage standard interfaces and service models to seamlessly integrate and interoperate across different Radio Access Network (RAN) vendor platforms. The ORAN architecture enables flexible placement of the sensing functions, in an Open Radio Unit (O-RU), Open Distributed Unit (O-DU) or a RAN Intelligent Controller (RIC) running as an embedded function and/or as software application (e.g., rApp or xApp) at the edge. The sensing function can even live in a software-defined radio (SDR) component within the O-RU. Each of these implementation options have a different set of fronthaul bandwidth, compute, performance and scalability tradeoffs. For purposes of this discussion, we show an exemplary implementation of the neural sensor within the O-RU with an associated xApp running in the Near Real Time RIC as illustrated in?Figure 1.
By sampling the received RF radio signal with a wideband receiver and digitizer in an Open Radio Unit (O-RU) and processing the resulting IQ (in-phase and quadrature) data stream with a specialized neural network, each O-RU can be effectively transformed into a powerful spectrum sensor. The neural network—an AI model pre-trained on a wide range of signals and conditions—ingests high-rate (Gbps) IQ data and detects and classifies the signals it sees in real time. The AI engine then outputs low bandwidth metadata (kbps) and a “semantic” detection of events, such as the sudden appearance of an airborne radar signal, carried in a standardized format via O1 or similar path as shown in?Figure 2. By exposing this signal detection event metadata to higher layers, it can be used by the distributed unit (DU) scheduler or RAN Intelligent Controller (RIC) to initiate appropriate interference mitigation or coexistence mechanisms.
?
The sampling parameters of the neural RF sensor can be tailored to optimize the use of available compute resources in the O-RU’s FPGA or other L1 processor, minimizing impact to BOM cost, power consumption and especially fronthaul bandwidth. While full-rate IQ data from RF receivers on the O-RU often requires costly 10 or 25 GbE for transport back to other network elements, RU-based detection and classification require only kbps of bandwidth to convey the resulting metadata containing detection events – offering orders of magnitude reduction in transport bandwidth and ease of scalability across the network to thousands of sensing locations.
With every O-RU a sensor, spatial coverage is more complete as compared to a limited ESC like sensor deployment so that incumbent users can be detected avoided. This enables a range of SMO/RIC- or DU scheduler-driven interference mitigation policies from dynamic blanking of 5G signals to beam steering or nulling to protect incumbent users while minimizing the impact on capacity of the 5G network.
Beyond Spectrum Sharing
With the potential to unlock large segments of new mid-band spectrum, interest in wideband spectrum sensing as an integral 5G (and 6G) network capability is growing. While DSS is emerging as the 5G killer application, the creation of network-wide RF awareness promises a range of multivendor applications that enhance RAN operations, efficiency and automation:
?OmniSIG: The First Name in AI Spectrum Sensing
Similar to early AI breakthroughs in image classification, DeepSig’s founders were the first to harness the power of deep learning to detect and classify RF signals with unprecedented accuracy. Built on a portfolio of fundamental patents for AI-native wireless communications, the company’s first software product, OmniSIG, has been widely vetted and adopted for defense and cyber security applications over the past 3-4 years and is seeing growing interest in 5G applications. Several visualizations of OmniSIG’s powerful spectrum sensing capabilities are shown in?Figure 3, illustrating OmniSIG tools for data curation and model training, model deployment, localization, and classification.
What is OmniSIG?
As shown in Figure 4 below, OmniSIG? is an industry-leading suite of machine learning (ML) software applications that adds RF awareness to a wide range of software-defined radios, radio systems, test and measurement devices and other wideband receivers. DeepSig’s customizable neural networks provide real-time identification, classification, and localization of known and unknown RF signals, automated alerting and reaction, and open standards-based descriptions of signal activity.
OmniSIG’s suite of software tools empower developers to create, deploy, and manage AI models for sensing of a wide range of communications and 5G systems—with unmatched performance and efficiency. OmniSIG Engine software can run on many NVIDIA GPUs, FPGAs, x86 or Arm processors and comes with native support for a wide range of software-defined radios (SDRs), ensuring broad compatibility and rapid integration with equipment across the wireless industry.
Now in its third generation, the OmniSIG Engine solution uses patented AI-driven signal detection and classification technology to enhance the reliability and security of mission-critical wireless communications for many large national defense and security organizations and commercial sensing applications. OmniSIG uses a customizable deep learning training pipeline that leverages neural networks highly tuned for RF sensing, using the state-of-the-art AI model architecture, to support a broad range of signal types, including LTE, 5G, Wi-Fi, Bluetooth, and more, with the flexibility to add custom signals. OmniSIG can detect and classify signals with high accuracy across a 400-500 MHz of bandwidth in as little as 1-2 milliseconds—1000x faster than many conventional methods.
For more information on OmniSIG, please go to:?https://www.deepsig.ai/spectrum-awareness/
References and Additional Resources:
Assistant Professor at MVSR Engineering College
2 周Thank you for sharing
Empowering brands to reach their full potential
1 个月Peter, thanks for sharing! How are you?
Great article on AI in spectrum sensing for 5G! ???? AI helps us better manage and protect radio frequencies in 5G networks.