Sub-Nyquist spectrum sensing and learning challenge

1 School of Computer Science and Electronic Engineering, University of Surrey, Guildford, Surrey GU2 7XH, UK 2 National Instruments Corporation (UK) Ltd, Newbury, Berkshire RG14 2PZ, UK 3 Electronic Media Services Ltd, Bordon, Hampshire GU35 0FJ, UK 4 Sony Research Center, Sony Corporation, Lund 221 88, Sweden 5 Institute for Networked Systems, RWTH Aachen University, Kackertstrasse 9, Aachen 52072, Germany 6 Faculty of Math & CS, Weizmann institute of Science, Rehovot 7610001, Israel 7 School of Science and Engineering, The Chinese University of Hong Kong, Shenzhen 518172, China 8 School of Electronic Engineering and Computer Science, Queen Mary University of London, London E1 4NS, UK

Still, the performance demands placed on sub-Nyquist spectrum sensing creates many different challenges, which comprise, but are not limited to, the following: • For compressive samplers, the necessary sampling rate to successfully reconstruct a sparse signal is determined by the actual sparsity order (the ratio of the occupied channel to the total sensing bandwidth) of the signal. On the other hand, spectrum reconstruction based on a greedy algorithm requires prior knowledge of spectrum sparsity as an input. However, due to the uncertainty in the environ-ment, the spectrum sparsity is always unknown and unpredictable. In practice, the sampling rate has to be chosen conservatively, according to the upper bound of the actual sparsity order instead of the real sparsity, which can be unnecessarily high, causing waste of sampling resources and computational burden, while the existing cross-validation algorithms are still compute-intensive [12,31]. • The spectrum dynamically changes over time. In practice, long-time statistical method should be avoided during the sensing stage to improve robustness, while the compressive recovery performance tends to deteriorate as the sampling window being shortened [32]. How to choose the suitable sampling windows is another challenge. • Algorithms to recover the spectrum from sub-Nyquist samples are often computationally intensive. It is desirable to spend as little time as possible on spectrum sensing to improve transmission efficiency and to reduce interference to PUs. • The transmission of the existing SUs should also be detected by the subsequent accessors. The coexistence of a large number of SUs can influence the spectrum sparsity, even beyond the capacity of the sub-Nyquist spectrum sensing device.
For stimulating novel approaches and designs on sub-Nyquist spectrum sensing and learning task. A challenge is issued with a reference sub-Nyquist algorithm, open data sets and awards up to 10,000 USD. It is hoped to promote relative research and facilitate the theory-to-practice process of promising ideas.

The challenge
Several Nyquist-rate time-domain data sets on baseband with GHz bandwidth are provided. In the meantime, basic MATLAB and LabVIEW codes of a sub-Nyquist sampling scheme with fundamental recovery algorithms are released for reference on the challenge website. The participants will be required to sense the spectrum from the given data sets as accurately as possible with a relatively lower average sampling rate at smaller computational cost. The participants will be judged on • The sensing ability and reconstruction accuracy of proposing approaches with the given data sets; • The robustness, complexity and real-time performance of proposing approaches working on real-world signal with our software-defined radio (SDR) test platform as shown in Section 6.
Team entrants are encouraged. Extra credits will be allocated to innovative methods.

Submission requirements
An overall sub-Nyquist spectrum sensing solution is requested, including the following two parts in general, with innovation or improvement in both or individual part.
• Sub-Nyquist Sampling architecture (include but not limited to analog-to-digital converter, modulated wideband converter and multicoset sampler, etc.); • Recovery & detection algorithms.

The documents for submission include
• MATLAB or Python code for processing the given data sets; • LabVIEW code for processing real-time data on the NI SDR test platform; • Algorithm and software design manual; • A concept paper demonstrating the sampling architecture and recovery & detection algorithms.

Challenge criteria
The submitted entries will be evaluated by the authors team and a few other experts in the field according to the criteria shown in Table 1.

The data sets
We provide data sets composed of digital samples of real wideband signal for participants to test their algorithms, the properties of the data sets are shown in Table 2. The data sets are composed of 500∼60000pts raw continuous baseband I/Q

Test platform
The submitted approaches will be tested on a hardware platform comprised of the NI mmWave SDR systems, used as the transmitter and receiver, respectively (Fig. 1). The transmitter and receiver have modular configurable hardware working at mmWave radio frequency centred at 28.5GHz with 2GHz bandwidth. The baseband signal consists of in-phase (I) and quadrature (Q) components with a frequency range of -1GHz to 1GHz. A single Nyquist ADC samples the baseband signal at a 3.072GSps rate at the receiver. Using NI LabVIEW development tools, the behaviour of the sub-Nyquist sampler can be simulated by pretreatments on Nyquist samples. The recovery algorithms implemented on the host controller process the real-time signal captured through the PCIe bus from the data acquisition card. An example implementation for reference is shown in [9]. Sample codes in MATLAB and data sets can be downloaded on the challenge website.

Challenge registration
The entrance for signing up for the challenge and submitting entries can be found at the Gbsense website. After registration, the data sets and the sample codes can be downloaded freely. The time nodes, awards and copyright rules are also announced 1) Participants may choose between MATLAB and Python, but LabVIEW code is necessary. on the website. Participants will win 10,000 USD for the first prize, 5,000 USD for the second prize, and 3,000 USD for the third prize. Petri Mähönen is currently a Full Professor and the Chair of Networked Systems with RWTH Aachen University, Germany. His current research interests include cognitive radio systems, embedded intelligence, and future wireless networks architectures, including millimeter-wave systems and technoeconomics especially from a regulatory perspective. He is also serving as an Editor for the IEEE Transactions on Wireless Communications. He is also corecipient of IEEE Jack Neubauer Memorial Award and received Telenor Research Prize for his work on spectrum related research. He is an expert on analysis and processing of audio, using a wide range of signal processing and machine learning methods. He led the first international data challenge on Detection and Classification of Acoustic Scenes and Events (DCASE 2013), and is a co-editor of the recent book on "Computational Analysis of Sound Scenes and Events" (Springer, 2018). He currently holds a 5-year EPSRC Fellowship on "AI for Sound", aiming to bring sound recognition technology "out of the lab" for the benefit of society.