Source-linked AI summary
Grant-Free Massive MTC-Enabled Massive MIMO: A Compressive Sensing Approach
Kamil Senel, Erik G. Larsson
TL;DR
The paper studies joint activity detection, channel estimation, and small-payload decoding in grant-free mMTC, where pilot collisions and CSI acquisition challenge coherent transmission. It combines CS-based detection with conventional estimation, introduces non-coherent pilot-embedded signaling and a modified AMP receiver, and reports improved coherent CSI and superior non-coherent performance over coherent transmission and original AMP.
Problem
Grant-free mMTC must jointly detect active devices, estimate channels, and decode small data payloads despite non-orthogonal pilot collisions and limited CSI resources.
Method
The paper combines CS-based device detection with conventional channel estimation and introduces pilot-embedded non-coherent transmission with a modified AMP receiver exploiting structured sparsity.
Results
The proposed approach improves channel estimates and coherent spectral efficiency, while non-coherent transmission significantly outperforms coherent transmission and modified AMP outperforms original AMP.
Takeaways & Limitations
Non-coherent transmission is more suitable for conveying small numbers of information bits in mMTC, especially for control signaling.
Abstract
from arXiv · showhide
A key challenge of massive MTC (mMTC), is the joint detection of device activity and decoding of data. The sparse characteristics of mMTC makes compressed sensing (CS) approaches a promising solution to the device detection problem. However, utilizing CS-based approaches for device detection along with channel estimation, and using the acquired estimates for coherent data transmission is suboptimal, especially when the goal is to convey only a few bits of data. First, we focus on the coherent transmission and demonstrate that it is possible to obtain more accurate channel state information by combining conventional estimators with CS-based techniques. Moreover, we illustrate that even simple power control techniques can enhance the device detection performance in mMTC setups. Second, we devise a new non-coherent transmission scheme for mMTC and specifically for grant-free random access. We design an algorithm that jointly detects device activity along with embedded information bits. The approach leverages elements from the approximate message passing (AMP) algorithm, and exploits the structured sparsity introduced by the non-coherent transmission scheme. Our analysis reveals that the proposed approach has superior performance compared to application of the original AMP approach.
I. INTRODUCTION
The paper addresses grant-free mMTC, where sparse activity, pilot collisions, and short packets make joint device detection, channel estimation, and data decoding difficult. It develops coherent and non-coherent approaches, including improved channel estimation, power control, and a structured AMP-based receiver.
- Motivation: mMTC traffic is uplink-driven, intermittent, and can contain only a few bits, making signaling overhead significant.These characteristics motivate communication methods tailored to sporadic short-packet transmissions.
- Problem: Grant-free random access avoids prior scheduling, but the massive device population prevents assigning orthogonal pilots to every device.Consequently, pilot collisions arise and conventional ALOHA-based access performs poorly when many devices access simultaneously.
- Prior approach: Compressed sensing detects active devices from concurrent transmissions of unique identifiers and can jointly estimate their channels.The paper also considers conventional channel estimation after CS-based device detection.
- Contributions: The paper combines CS-based detection with conventional estimators to obtain more accurate CSI and investigates power control for improving device detection.It derives a coherent-transmission spectral-efficiency result for the combined scheme.
- Results: Under mMTC setups, non-coherent transmission significantly outperforms coherent transmission and is more suitable for conveying small numbers of information bits.The paper compares the two approaches specifically when device detection, channel estimation, and data decoding use non-orthogonal pilots.
- Contributions: The proposed non-coherent scheme embeds information bits in pilot sequences and jointly detects active devices and their associated bits without prior channel or activity information.The receiver uses a modified AMP approach that exploits the structured sparsity introduced by the scheme.
II. SYSTEM SETUP
The system models uplink mMTC with a large-antenna base station serving many intermittently active devices over coherence intervals. Non-orthogonal pilots create a row-sparse channel-recovery problem, while coherent transmission divides each interval between channel estimation and data transmission.
- System model: A single base station with M antennas serves N devices, with only a subset active in each coherence interval.The channel is constant and frequency-flat for τ samples; large-scale fading is known, while small-scale fading is estimated per interval.
- Coherent transmission: Coherent transmission uses τp pilot symbols followed by τ − τp data symbols, requiring the base station to detect activity, estimate channels, and decode data.Orthogonal pilots would require τp ≥ N and are infeasible for large mMTC populations.
- Pilot design: The system therefore uses non-orthogonal pilot sequences generated from an i.i.d. Bernoulli distribution.Bernoulli pilots have a non-zero collision probability because only finitely many unique sequences exist.
- Pilot design: For N = 200 devices and pilot length τp = 20, the pilot collision probability is approximately 10^-8.The base station knows the device-associated pilot sequences, which can be generated using device identifiers as pseudorandom seeds.
- Detection model: The effective channel matrix is row-sparse because inactive devices correspond to zero rows, so activity detection reduces to finding its non-zero rows.The received composite signal includes device activity indicators, additive white Gaussian noise, and common uplink transmission power before power-control analysis.
- Design motivation: Short coherence intervals and intermittent traffic make allocating orthogonal pilots and spending resources on CSI potentially unsuitable for mMTC.Higher frequency bands and mobility can further reduce the coherence interval, motivating alternative data-transmission approaches.
III. REVIEW OF APPROXIMATE MESSAGE PASSING
The paper reviews AMP as a low-complexity compressed-sensing method for recovering the row-sparse effective channel matrix from noisy pilot observations. AMP uses iterative denoising with an Onsager correction and admits asymptotic state-evolution analysis.
- Problem formulation: With multiple antennas, active-device detection is formulated as a multiple-measurement-vector compressed-sensing reconstruction problem.The single-antenna case instead corresponds to single-measurement-vector reconstruction.
- AMP method: AMP is used to recover the sparse effective channel matrix and thereby identify active devices from the received observations and known pilot sequences.The method is presented as a low-complexity compressed-sensing algorithm.
- AMP method: Each AMP iteration updates an estimate through a denoising function and updates the residual using the derivative of that function.The derivative-dependent correction is the Onsager term, which substantially improves iterative performance.
- State evolution: In the asymptotic regime where τp, K, and N grow with fixed ratios, AMP behavior is described by state-evolution equations.The analysis models active-device channel vectors with a Gaussian distribution and inactive-device channels with a point mass at zero.
SIMULATION PARAMETERS
The simulations evaluate AMP-based device detection under varying pilot lengths, antenna counts, and pilot distributions. Results show that longer pilots, more antennas, and Bernoulli pilots improve detection, with diminishing antenna returns.
- AMP setup: AMP state evolution models device-wise denoising for multiuser detection under stated asymptotic assumptions.The denoising function is MMSE for the equivalent system and uses likelihood-ratio thresholding for activity detection.
- Antenna count: Increasing the number of BS antennas significantly improves AMP user detection, but the gains gradually saturate as M increases.The antenna count should therefore not be treated as an absolute substitute for pilot sequence length.
- Pilot distribution: Bernoulli pilot sequences outperform Gaussian pilots for device activity detection, with the performance gap becoming larger at longer pilot lengths.Bernoulli sequences are also motivated by their finite alphabet and practical implementation.
B. Asymptotic Analysis
The asymptotic analysis characterizes AMP device detection through state evolution and shows that miss-detection and false-alarm probabilities vanish as the antenna count grows. Simulations with Gaussian and Bernoulli pilots follow this predicted improvement, with Bernoulli pilots performing better.
- Asymptotic analysis: State evolution decouples AMP estimation across devices and supports theoretical analysis of device detection in the asymptotic region.The analysis assumes state evolution holds in this region and uses it to characterize AMP performance.
- Scope and assumptions: The characterization of sensing matrices for which state evolution holds remains an open problem, despite numerical evidence for a broader matrix class.The analysis therefore assumes state evolution in the asymptotic region.
- Asymptotic analysis: Under uncorrelated channels across antennas, activity detection by thresholding the AMP estimate has miss-detection and false-alarm probabilities that go to zero as M →∞.This result holds for thresholds satisfying the lemma’s stated condition.
- Numerical verification: Bernoulli sequences provide better detection performance than Gaussian sequences in the asymptotic comparison.The same ordering is reported for the large-device simulation as M varies.
IV. POWER CONTROL
The paper evaluates simple power control for mMTC and combines AMP-based device detection with MMSE channel estimation for coherent transmission. Statistical channel inversion reduces interference and improves detection, while MMSE can provide more accurate channel estimates after activity detection.
- Power control: Statistical channel inversion (SCI) scales transmission powers inversely with large-scale fading coefficients, with the weakest device transmitting at maximum power.Devices below the minimum coefficient βmin cannot access the network within their available power budget.
- Power control: SCI consumes less total power than no power control (NPC), resulting in lower total interference.Under NPC, every device transmits with maximum power.
- Power control: Even a simple power control policy improves device detection performance, with a larger difference as the number of antennas increases.The comparison is between NPC and SCI.
- Coherent transmission: For coherent transmission, AMP first detects active devices, after which MMSE estimates their channels using the detected active set.The MMSE estimator achieves the true MMSE when pilot sequences and large-scale fading coefficients are known at the base station.
- Coherent transmission: MMSE after AMP detection enables higher achievable rates than using the AMP channel estimates alone, while adding less complexity than one AMP iteration.The analysis assumes perfect device detection to isolate channel-estimation accuracy, with imperfect detection examined later.
- Coherent transmission: Around τp = 20, the rate difference between perfect and imperfect device detection becomes negligible, and the difference between AMP and MMSE estimates also vanishes.Both approaches approach the perfect-CSI case as pilot length increases, while perfect-CSI rate decreases because of pilot overhead.
VI. NON-COHERENT TRANSMISSION
The non-coherent scheme embeds information bits in pilot-sequence selection, allowing devices to transmit data without separate data symbols. Modified AMP exploits the resulting structured sparsity to jointly detect activity and embedded bits.
- Transmission scheme: To transmit r bits non-coherently, each device receives 2^r distinct pilot sequences and transmits one sequence selected by its information bits.The information is embedded directly in the pilot sequence.
- Transmission scheme: Because no additional data symbols are allocated, the entire coherence interval can be used for pilot sequences.This contrasts with coherent transmission, which requires explicit channel estimates and separate data signaling.
- Structured sparsity: The base station considers N2^r pilot-sequence candidates, but the number of active users remains unchanged because each device transmits exactly one sequence.The one-sequence constraint creates structural sparsity in the expanded signal matrix.
- Detection algorithm: Applying unmodified AMP treats pilot sequences as belonging to fictitious independent devices and is strictly suboptimal because it ignores the structure of the expanded signal.The proposed modified AMP algorithm extends the approach to general r.
A. Algorithm Description
The modified AMP algorithm uses the structured sparsity created by non-coherent pilot selection: for each device, at most one corresponding row can be non-zero. Its performance exceeds original AMP when detecting embedded bits with device activity.
- Algorithm Description: Assigning multiple pilot sequences to each device increases the number of candidate rows while preserving device-level structural sparsity.Rows corresponding to alternative sequences of the same device cannot be simultaneously non-zero.
- Algorithm Description: The sequence likelihood fraction (SLF) coefficient measures the proportional likelihood of a given sequence and provides proportional thresholding.A sigmoid soft-thresholding function sharpens the decision, with parameter c controlling transition sharpness.
- Limitations: The modified denoiser is Lipschitz-continuous, but state-evolution validity remains unclear because the unmodified Bernoulli case is verified only numerically.The asymptotic behavior of AMP for non-Gaussian sensing-matrix distributions remains an open problem.
- Algorithm Description: M-AMP is specifically designed for non-coherent transmission, where only one row per device may be non-zero because a device cannot transmit both pilot sequences concurrently.The detector uses this constraint rather than treating every candidate sequence as an independent device.
- Performance comparison: The performance gap between M-AMP and original AMP becomes more significant as pilot length increases.The comparison uses identical iteration counts and pilot-sequence lengths across algorithms.
- Scaling behavior: When N, τp, and K increase with their ratios fixed, all approaches outperform the corresponding 100-user case while exhibiting similar scaling behavior.This scaling is desirable for mMTC scenarios with many devices.
B. Coherent versus Non-Coherent Transmission
The paper compares coherent and non-coherent transmission for short mMTC packets using probability of error. Non-coherent transmission performs better and scales better with coherence length for one bit, while its advantage narrows as the payload grows.
- B. Coherent versus Non-Coherent Transmission: The comparison uses probability of error because the target applications transmit only a few data bits.The single-bit experiment varies coherence-interval length and repetition-code length.
- Coherent transmission: Coherent transmission first detects active devices and estimates channels with AMP, then sends one BPSK bit using repetition coding.Most of the coherence interval is used for pilots, except for the repetition-coded information bit.
- Single-bit transmission: The best coherent single-bit performance uses a length-11 repetition code, while lengths 15 and 19 perform similarly.The setup includes M = 20 antennas and N = 100 devices.
- Single-bit transmission: Non-coherent transmission outperforms coherent transmission and scales better with coherence-interval length for a single information bit.The comparison is reported in terms of probability of error.
VII. CONCLUSION
The paper addresses joint device detection and data transmission in mMTC using AMP-based detection, power control, improved channel estimation, and a proposed non-coherent scheme. The proposed methods improve detection, channel estimation, spectral efficiency, and performance relative to the corresponding baselines.
- A simple power-control technique based only on large-scale coefficients enhances device-detection performance.
- MMSE estimation after active-device detection produces more accurate channel estimates than AMP estimates and increases coherent-transmission spectral efficiency.
- The non-coherent scheme embeds information bits in each device’s pilot-sequence choice, mapping r bits onto 2^r possible pilots.
- M-AMP exploits the structured sparsity created by the non-coherent scheme to detect devices and embedded bits.
- M-AMP outperforms original AMP for the non-coherent scheme and scales better with the number of devices.
- Non-coherent transmission significantly outperforms coherent transmission and may be useful for future mMTC communication, particularly control signaling.
APPENDIX A
The appendix develops asymptotic analyses for miss-detection and false-alarm probabilities using Gamma-function representations and large-system approximations. It derives threshold conditions under which both error probabilities vanish as the number of base-station antennas grows.
- The asymptotic regime lets τp, K, and N grow while their ratios remain fixed.
- The analysis represents miss-detection and false-alarm probabilities using Gamma-function expressions.
- As M →∞, both PrMD(M, ζ) and PrFA(M, ζ) approach zero for any choice of the analyzed parameter.
- Thresholds satisfying the derived interval yield perfect detection in the asymptotic region.