Source-linked AI summary
Compressive Sensing Techniques for Next-Generation Wireless Communications
Zhen Gao, Linglong Dai, Shuangfeng Han, I Chih-Lin, Zhaocheng Wang, Lajos Hanzo
TL;DR
Next-generation wireless systems face high sampling and resource demands as bandwidth, antennas, connectivity, and network density grow, while many relevant signals and channels are sparse. The paper surveys compressive-sensing models and recovery techniques, then examines their use across 5G technologies and concludes with open practical research directions.
Problem
5G’s wider bandwidth, larger antenna arrays, and dense connectivity create sampling, overhead, complexity, and interference-management challenges that exploitations of sparsity may address.
Method
The paper surveys compressive-sensing theory, models, recovery algorithms, and applications that exploit sparsity in 5G channels, signals, spectrum, interference, and traffic.
Results
The survey identifies CS-based opportunities for reducing channel-sounding, estimation, detection, hardware, power, interference-management, random-access, and traffic-prediction overheads across 5G techniques.
Takeaways & Limitations
CS is presented as a promising framework for exploiting multifold sparsity in next-generation wireless systems and motivating research on practical algorithms and hardware-compatible applications.
Abstract
from arXiv · showhide
A range of efficient wireless processes and enabling techniques are put under a magnifier glass in the quest for exploring different manifestations of correlated processes, where sub-Nyquist sampling may be invoked as an explicit benefit of having a sparse transform-domain representation. For example, wide-band next-generation systems require a high Nyquist-sampling rate, but the channel impulse response (CIR) will be very sparse at the high Nyquist frequency, given the low number of reflected propagation paths. This motivates the employment of compressive sensing based processing techniques for frugally exploiting both the limited radio resources and the network infrastructure as efficiently as possible. A diverse range of sophisticated compressed sampling techniques is surveyed and we conclude with a variety of promising research ideas related to large-scale antenna arrays, non-orthogonal multiple access (NOMA), and ultra-dense network (UDN) solutions, just to name a few.
I. INTRODUCTION
5G must accommodate sharply higher bandwidth, antenna counts, base-station density, and user connectivity than earlier cellular generations. The paper surveys compressive sensing as a way to exploit sparsity across these techniques and identifies research directions for practical systems.
- Motivation: 5G introduces at least 100 MHz bandwidth, hundreds of antennas, and ultra-dense base-station deployments, making direct reliance on conventional Nyquist sampling challenging.Earlier systems used up to 20 MHz bandwidth and eight antennas in 4G.
- Paper scope: The paper organizes compressive-sensing opportunities around key 5G techniques and discusses their associated opportunities and challenges.The treatment covers three technical directions and introduces CS concepts, models, and recovery algorithms.
- Sparsity exploitation: CIR sparsity, spatial-modulation signal sparsity, and NOMA codeword sparsity can reduce channel-sounding overhead and signal-detection complexity.These applications target massive MIMO, massive SM-MIMO, and NOMA systems.
- Sparsity exploitation: Sparse spectrum occupation and UWB signal sparsity can reduce hardware cost and power consumption, while mmWave CIR sparsity can improve precoding and reduce estimation overhead.The paper links these opportunities to cognitive radio, UWB, and mmWave communications.
- Sparsity exploitation: Sparsity in interference, coordinated transmission, random access, and traffic load can reduce overheads in ultra-dense networks.The paper highlights inter-cell-interference mitigation, CoMP, large-scale random access, and traffic prediction.
II. KEY TECHNICAL DIRECTIONS IN 5G
The paper frames 5G capacity growth through increased spectral efficiency, larger transmission bandwidth, and better spectrum reuse. Each direction brings specific channel-estimation, waveform, hardware, or interference-management challenges.
- Three directions: Network capacity can grow through increased spectral efficiency, larger transmission bandwidth, and better spectrum reuse.The paper associates these dimensions with more channels, spectrum sharing or extension, and more cells per area.
- Increased spectral efficiency: Massive multi-antenna spatial multiplexing can boost capacity, but channel estimation and massive SM-MIMO signal detection remain challenging.The paper identifies these as unresolved issues accompanying increased spectral efficiency.
- Increased spectral efficiency: NOMA can theoretically support more users than conventional OMA under limited radio resources, while sparse-codeword design approaching NOMA capacity remains open.The open problem concerns the optimal design of sparse codewords.
- Larger transmission bandwidth: Cognitive radio and UWB can enlarge transmission bandwidth while coexisting with licensed services through spectrum sharing, making sub-Nyquist sampling important.The paper also identifies mmWave communications as a candidate for high data rates through wider bandwidth.
- Better spectrum reuse: Small cells improve area spectral efficiency, but interference mitigation, CoMP transmission/reception, and massive random access remain substantial challenges.The cited area spectral-efficiency unit is bits/sec/Hz/km2.
III. COMPRESSIVE SENSING THEORY
Compressive sensing reconstructs sparse or transform-sparse signals from fewer measurements by combining sparse transformation, measurement compression, and recovery algorithms. The paper surveys CS models and algorithm families relevant to wireless applications.
- CS foundations: Correlated real-world signals often have fewer effective degrees of freedom than their sampled dimensions and may be sparse in the frequency domain.Examples include voiced speech, adjacent video pixels, and correlated fading-channel envelopes.
- CS foundations: The standard CS model represents measurements as y = Φx, with m ≪ n measurements, and can express transform-domain sparsity as x = Ψs.The sparse signal has k ≪ n non-zero elements, and recovery seeks x from y and Φ.
- CS models: Models (2)–(4) support multiple sparse signals, block sparsity, and identical or partially common sparsity patterns for more reliable recovery.These models exploit application-specific sparse properties.
- CS foundations: CS has three fundamental elements: sparse transformation, dimension-reducing measurement design, and sparse-signal recovery.Measurement information loss can be quantified using coherence or the restricted isometry property.
- Recovery algorithms: Convex-relaxation algorithms use optimization and few measurements but can be complex, whereas greedy methods are faster and lower-complexity but lose performance when signals are not very sparse.Representative methods include BP/BPDN, OMP, CoSaMP, and SP.
- Recovery algorithms: Bayesian methods infer sparse signals using sparse priors, while algorithms for models with structured sparsity remain promising future candidates.Examples include sparse Bayesian learning, approximate message passing, SOMP, and group-sparse Bayesian CS.
- Wireless applications: CS prototypes have been reported for MIMO radar, cognitive radio, and UWB, but practical 5G deployment still requires further investigation.The paper highlights reduced complexity, increased reliability, and hardware compatibility as future directions.
IV. HIGHER SPECTRAL EFFICIENCY
Massive MIMO, massive SM-MIMO, and NOMA are presented as promising routes to higher 5G spectral efficiency. The section focuses on exploiting their inherent sparsity.
- Higher spectral efficiency: Massive MIMO, massive SM-MIMO, and NOMA constitute promising candidates for increasing 5G spectral efficiency through sparsity exploitation.The paper introduces this as its first technical direction.
A. Massive MIMO Schemes
Massive MIMO can serve multiple users with hundreds of base-station antennas, but FDD channel estimation and feedback impose prohibitive overhead. Channel sparsity motivates compressive-sensing approaches, although compatibility with existing Nyquist-rate systems remains unresolved.
- Hundreds of base-station antennas enable simultaneous multi-user service with improved spectral and energy efficiency.
- FDD massive MIMO requires estimating hundreds of downlink antenna-pair channels and feeding their channel state information back to the base station.This creates very high pilot and feedback overhead, while analog CSI feedback is described as unaffordable.
- Limited significant scatterers and strong spatial correlation make massive MIMO channels sparse in delay, angular, or both domains.Delay-domain channels can contain far fewer energy-dominant paths than total channel impulse-response taps, supporting structured block sparsity across antenna pairs.
- Compressive sensing is expected to reduce channel-estimation and feedback overhead by exploiting massive MIMO channel sparsity.
- Sub-Nyquist-tailored pilots may require further research to remain compatible with existing classic-Nyquist systems.
B. Massive SM-MIMO Schemes
Massive SM-MIMO activates only a fraction of antennas, producing sparse signals that compressive sensing can use for detection. This reduces RF-chain demands and detection complexity, but channel reconstruction remains challenging.
- Massive SM-MIMO uses hundreds of antennas while activating a much smaller number of RF chains and antennas for each transmission slot.Only a small fraction of antennas carries classically modulated signals at a time.
- Fig. 2 highlights the sparsity of SM signals in massive SM-MIMO systems.
- Spatial modulation improves performance but provides no transmit-diversity gain, a problem that can be mitigated by activating a limited fraction of antennas.
- Sparse downlink and aggregated uplink SM signals create large under-determined detection problems that can be formulated with a standard compressive-sensing model.The sparse signal is mapped through the MIMO channel matrix to the received signal.
- SM-signal sparsity can reduce receiver computational complexity for signal detection.
C. Sparse Codewords in NOMA systems
NOMA can support more users through non-orthogonal resources, while SCMA uses sparse user-specific codewords for multi-user separation. Compressive sensing is proposed for improving the associated performance–complexity trade-off.
- NOMA can potentially support more users than orthogonal multiple access by using non-orthogonal resources, typically with increased receiver complexity.
- SCMA assigns each user a unique non-orthogonal spreading sequence, and each codeword is sparse and represents a transmission layer.
- Sparse SCMA codewords allow the base station to distinguish multiple uplink users over non-orthogonal resources.The signal-detection problem can be formulated as a signal-separation model, with message passing providing near-maximum-likelihood performance at low complexity.
- Compressive sensing may guide optimal SCMA codeword and multi-user-detector design to improve the performance-versus-complexity trade-off.
A. Cognitive Radio
Cognitive radio exploits sparsely occupied licensed spectrum, but broad spectrum sensing is difficult at conventional Nyquist rates. Compressive sensing enables sub-Nyquist sensing and lower-speed ADCs, while compressed-domain demodulation is possible when transmission parameters are known.
- Large portions of licensed spectrum remain under-utilized, motivating cognitive radio sensing of spectrum holes for secondary-user access.
- Compressive spectrum sensing uses sub-Nyquist sampling to sense broad spectrum despite the high conventional Nyquist sampling rate.
- CS-based cognitive radio can replace high-speed ADC requirements with low-speed analog-to-digital converters.
- Xampling can demodulate compressed received signals when frame structure and modulation modes are known.
- UWB signals require high Nyquist sampling rates because of their GHz bandwidth, increasing receiver ADC power consumption and hardware cost.
- UWB’s intrinsic time-domain sparsity permits sub-Nyquist recovery, but extracting complete analog-signal information from compressed measurements remains challenging.Receivers seeking only conveyed information may instead process compressed measurements directly without reconstructing the waveform.
C. Millimeter-Wave Communications
Millimeter-wave communications offer wide bandwidth but face spatially sparse channels, low pre-beamforming SNR, limited RF chains, and challenging channel estimation and precoding. Compressive sensing exploits channel sparsity to reduce training overhead and simplify hybrid precoding, although broadband extensions remain under investigation.
- mmWave communications provide wider bandwidth for high data rates but exhibit spatially sparse channels, low pre-beamforming SNR, and fewer RF chains.
- Full-digital precoding is impractical under hardware constraints because each antenna requires a dedicated RF chain, increasing cost and power consumption.
- Hybrid precoding can be formulated as sparse signal recovery and efficiently solved with a modified OMP algorithm.
- CS-based channel estimation exploits angular-domain sparsity and low-rank multipath structure to reconstruct mmWave channels with fewer RF chains and reduced training overhead.
- Existing CS methods reduce hybrid-precoding complexity and channel-estimation overhead, but extending narrow-band solutions to broadband mmWave MIMO remains under investigation.
A. BSs Identification
Ultra-dense networks create interference and massive-access challenges because many base stations and potential users share limited orthogonal resources, while only a small subset is typically active or interfering. Compressive sensing identifies those sparse participants and supports interference mitigation and random access with reduced overhead, but robust signal design remains open.
- Limited orthogonal resources make inter-cell-interference mitigation in ultra-dense small cells an open challenge.
- Although many base stations are available, each user is usually interfered by only a small number, creating a sparse identification problem.
- Non-orthogonal training signals let users detect interfering-base-station identities and channel information with small overhead.
- Block-sparsity and MMV models can use observations across antennas or frames to improve interfering-base-station identification.
- In massive random access, the active-user fraction remains small despite many potential users, motivating CS-based access methods.
- Uplink random access is harder than downlink base-station identification because there are many more users and cooperative feedback compression is challenging.
C. Traffic Estimation and Prediction for Energy-Efficient Dense Networks
Dense-network traffic estimation supports dynamic radio-access management and energy efficiency, but Nyquist-based measurement can impose substantial resource demands. Strong spatio-temporal traffic correlation creates a low-rank structure that compressive sensing can reconstruct from sub-Nyquist or partial measurements, while practical CS deployment remains an open direction.
- Traffic-load estimation is needed to dynamically manage radio access and improve energy efficiency, but Nyquist-based estimation requires substantial measurements, storage, feedback, and energy.
- Daily periodicity and human-activity-driven spatial variation produce strong spatio-temporal traffic correlation and a low-rank traffic-indicator matrix.
- Sub-Nyquist sampling and spatio-temporal Kronecker compressive sensing can reconstruct complete traffic matrices from partial traffic data.
- Exploiting low rank reduces measurements and feedback required for traffic prediction or transmission to a fusion center.
- Compressive sensing addresses large overhead, complexity, cost, and power challenges arising in wideband, antenna-rich, and ultra-dense 5G systems.
- Theoretical CS research has progressed substantially, but practical applications still require lower complexity, higher reliability, and hardware compatibility.