A Deep-learning-based Joint Inference for Secure Spatial Modulation Receiver

Feng Shu; Guiyang Xia; Jiangzhou Wang; Jun Li; Lin Liu; Shi Jin; Xiaoyu Liu; Yumeng Zhang

arxiv: 1907.02215 · v1 · pith:J7IR6MHAnew · submitted 2019-07-04 · 📡 eess.SP · cs.IT· math.IT

A Deep-learning-based Joint Inference for Secure Spatial Modulation Receiver

Feng Shu , Lin Liu , Yumeng Zhang , Guiyang Xia , Xiaoyu Liu , Jun Li , Shi Jin , Jiangzhou Wang This is my paper

Pith reviewed 2026-05-25 09:29 UTC · model grok-4.3

classification 📡 eess.SP cs.ITmath.IT

keywords secure spatial modulationdeep neural networkjoint detectionmaximum likelihood detectionbit error rateartificial noiseantenna selectionpower allocation

0 comments

The pith

A deep neural network with redesigned layers jointly infers antenna index and symbol in secure spatial modulation, reaching near-maximum-likelihood bit error rates at far lower complexity.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes a deep neural network detector for secure spatial modulation receivers that carry information in both the active transmit antenna index and the amplitude-phase modulated symbol. Each layer is restructured specifically to optimize the joint inference of these two heterogeneous data types rather than treating them separately. Simulation results indicate the method delivers bit error rates about 3 dB better than a standard DNN structure and close to optimal maximum-likelihood detection in low-to-medium signal-to-noise ratios, while requiring substantially less computation than ML. The approach is presented alongside transmitter techniques of antenna selection, power allocation, and artificial-noise projection to realize secure transmission. A reader would care because detection complexity has been a practical barrier to deploying this energy-efficient, physically secure modulation format.

Core claim

The central claim is that a deep neural network whose layers are redesigned to jointly infer the transmit antenna index and the constellation symbol point achieves bit error rate performance close to maximum-likelihood detection in the low and medium SNR regions for secure spatial modulation, while exhibiting far lower complexity than ML detection.

What carries the argument

Redesigned DNN layers that jointly optimize inference of two distinct information types: transmit antenna index and signal constellation point.

If this is right

Secure spatial modulation can be made practically viable by pairing the low-complexity DNN detector with transmit antenna selection, power allocation, and artificial noise projection.
The joint-inference redesign yields a 3 dB BER gain over conventional DNN structures while remaining close to ML performance.
The lower computational cost relative to ML enables real-time operation in resource-constrained receivers.
The same transmitter techniques can be combined to turn spatial modulation into a genuinely secure modulation scheme.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If the layer redesign proves robust, similar joint-inference networks could be applied to other MIMO detection tasks that involve heterogeneous symbol types.
Deployment would likely require periodic retraining or domain adaptation when channel statistics change beyond the training distribution.
The reported complexity reduction could be quantified further by measuring floating-point operations or latency on embedded hardware.

Load-bearing premise

A neural network trained to jointly detect antenna index and symbol will generalize from its training data to real wireless channels without large performance loss from mismatch between the two information types.

What would settle it

An experiment or simulation in which the proposed DNN's bit error rate deviates by more than 1 dB from ML detection across the low-to-medium SNR range or shows higher complexity than claimed under identical channel conditions.

Figures

Figures reproduced from arXiv: 1907.02215 by Feng Shu, Guiyang Xia, Jiangzhou Wang, Jun Li, Lin Liu, Shi Jin, Xiaoyu Liu, Yumeng Zhang.

**Figure 1.** Figure 1: Schematic diagram of secure SM network. fields such as computer vision, natural language processing, speech recognition, etc., and has achieved great success. Due to the new features of future communication, such as complex scenes with unknown channel models, high-speed and accurate processing requirements, many scholars have introduced DL into the physical layer of wireless communication [9]. In the physi… view at source ↗

**Figure 3.** Figure 3: Comparison of SR performance of various PA strategie [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

**Figure 2.** Figure 2: Comparison of SR performance of various TAS methods. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 4.** Figure 4: Comparison of SR of various PA strategies. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗

**Figure 5.** Figure 5: Proposed novel DNN structure for joint inference of a [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗

**Figure 6.** Figure 6: BER performance comparison among proposed DNN, join [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗

read the original abstract

As a green and secure wireless transmission way, secure spatial modulation (SM) is becoming a hot research area. Its basic idea is to exploit both the index of activated transmit antenna and amplitude phase modulation (APM) signal to carry messages, improve security, and save energy. In this paper, we reviewed its crucial techniques: transmit antenna selection (TAS), artificial noise (AN) projection, power allocation (PA), and joint detection at desired receiver. To achieve the optimal performance of maximum likelihood (ML) detector, a deep-neural-network (DNN) joint detector is proposed to jointly infer the index of transmit antenna and signal constellation point with a lower-complexity. Here, each layer of DNN is redesigned to optimize the joint inference performance of two distinct types of information: transmit antenna index and signal constellation point. Simulation results show that the proposed DNN method performs 3dB better than the conventional DNN structure and is close to ML detection in the low and medium signal-to-noise ratio regions in terms of the bit error rate (BER) performance, but its complexity is far lower-complexity compared to ML. Finally, three key techniques TAS, PA, and AN projection at transmitter can be combined to make SM a true secure modulation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper offers an incremental DNN layer redesign for joint antenna-symbol detection in secure SM but the 3 dB gain claim rests on high-level simulations without training or evaluation details.

read the letter

The core move is redesigning each DNN layer to handle joint inference of transmit antenna index and constellation symbol in a secure spatial modulation receiver. This produces the reported 3 dB BER improvement over a conventional DNN and performance close to ML at low-to-medium SNR with far lower complexity. The abstract also folds in the usual transmitter-side pieces (TAS, AN projection, power allocation) to close the secure SM loop. That combination is the main thing the work adds; it is a direct extension of existing DNN detectors rather than a new framework. The layer redesign itself is a reasonable engineering step for mixing two heterogeneous outputs in one network. The simulations are presented as the supporting evidence, and the complexity reduction versus ML is the clearest practical point. The main weakness is the absence of any description of training data, channel model, hyper-parameter search, or statistical checks on the 3 dB figure. Without those, the performance numbers cannot be assessed for robustness or sensitivity to the training distribution. The concern about loss or gradient imbalance between the categorical antenna task and the symbol task is also unaddressed in the given text, so it remains possible that one output dominates and the reported gain is setup-specific. This paper is aimed at the small group already working on DNN receivers for physical-layer security in MIMO. It does not contain enough verifiable detail to stand on its own, but the underlying idea is coherent enough that a serious editor could send it for review once the methods and results sections are expanded with the missing experimental information.

Referee Report

3 major / 2 minor

Summary. The paper reviews key techniques in secure spatial modulation (TAS, AN projection, PA) and proposes a DNN-based joint detector that redesigns each layer to simultaneously infer the transmit antenna index and the APM constellation symbol. It claims this architecture yields a 3 dB BER improvement over a conventional DNN detector, approaches ML performance at low-to-medium SNR, and has far lower complexity than ML; the final sentence asserts that combining TAS, PA, and AN makes SM truly secure.

Significance. If the performance claims are reproducible, the work would supply a concrete low-complexity alternative to ML detection for secure SM, which is relevant for energy-efficient and physically secure wireless links. The explicit redesign of layers for heterogeneous outputs (categorical antenna index vs. geometric constellation symbols) is a targeted architectural contribution, though its effectiveness remains to be verified.

major comments (3)

[Simulation results / abstract] The central performance claim (3 dB gain and near-ML BER) rests entirely on simulation results whose training protocol is not described: no information is given on the size or distribution of the training set, the channel model used for training versus testing, the loss function that combines the two heterogeneous tasks, or any hyper-parameter search procedure. Without these details the reported gain cannot be assessed for reproducibility or overfitting.
[Proposed DNN joint detector (architecture description)] The manuscript does not specify how the joint loss or back-propagation path balances the antenna-index classification task against the constellation-symbol detection task. Because the two outputs have different cardinalities and geometries, an imbalance could allow one task to dominate parameter updates, rendering the claimed superiority of the redesigned layers an artifact of the particular training distribution rather than a general property.
[Simulation results] No statistical significance measures (number of Monte-Carlo trials, confidence intervals, or variance across random seeds) accompany the BER curves that support the 3 dB and “close to ML” statements, making it impossible to judge whether the observed differences are load-bearing or within simulation noise.

minor comments (2)

[Abstract] Abstract contains the repeated phrase “far lower-complexity”; correct to “far lower complexity”.
[Abstract / conclusion] The final sentence asserts that TAS, PA, and AN projection “can be combined to make SM a true secure modulation,” yet the manuscript provides neither analysis nor simulation of the combined scheme; this claim should be either substantiated or removed.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thorough review and constructive comments on our manuscript regarding the DNN-based joint detector for secure spatial modulation. We address each major comment point-by-point below, agreeing that additional details will improve reproducibility and clarity. We will revise the manuscript accordingly.

read point-by-point responses

Referee: [Simulation results / abstract] The central performance claim (3 dB gain and near-ML BER) rests entirely on simulation results whose training protocol is not described: no information is given on the size or distribution of the training set, the channel model used for training versus testing, the loss function that combines the two heterogeneous tasks, or any hyper-parameter search procedure. Without these details the reported gain cannot be assessed for reproducibility or overfitting.

Authors: We agree these details are essential for assessing reproducibility. In the revised manuscript, we will add a dedicated subsection on the simulation setup, specifying the training set size and generation (e.g., random samples from the secure SM constellation and antenna indices), the channel model (identical Rayleigh fading for training and testing), the joint loss function (weighted combination of cross-entropy for antenna classification and MSE for symbol detection), and the hyperparameter search (grid search over learning rate, epochs, and layer sizes). revision: yes
Referee: [Proposed DNN joint detector (architecture description)] The manuscript does not specify how the joint loss or back-propagation path balances the antenna-index classification task against the constellation-symbol detection task. Because the two outputs have different cardinalities and geometries, an imbalance could allow one task to dominate parameter updates, rendering the claimed superiority of the redesigned layers an artifact of the particular training distribution rather than a general property.

Authors: We acknowledge the need to clarify loss balancing for the heterogeneous tasks. The revision will explicitly describe the joint loss formulation, including any weighting coefficients to balance the antenna-index (categorical) and constellation-symbol (geometric) tasks, and detail how back-propagation flows through the shared redesigned layers. We will also note any empirical tuning to avoid dominance by one task. revision: yes
Referee: [Simulation results] No statistical significance measures (number of Monte-Carlo trials, confidence intervals, or variance across random seeds) accompany the BER curves that support the 3 dB and “close to ML” statements, making it impossible to judge whether the observed differences are load-bearing or within simulation noise.

Authors: We agree that statistical measures would strengthen the results. The revised manuscript will report the number of Monte-Carlo trials per BER point and include confidence intervals or variance estimates across random seeds where applicable, to confirm the 3 dB gain and proximity to ML are not due to simulation noise. revision: yes

Circularity Check

0 steps flagged

No circularity: performance claims rest on independent simulations

full rationale

The paper presents a DNN architecture with redesigned layers for joint detection of antenna index and constellation symbols, then reports BER curves from Monte-Carlo simulations that compare the proposed network against a conventional DNN and ML detection. No derivation, uniqueness theorem, or parameter fit is invoked whose output is algebraically identical to its input; the central performance numbers (3 dB gain, proximity to ML at low/medium SNR) are generated by running the trained network on held-out channel realizations and are therefore falsifiable outside any self-referential loop. Self-citations are absent from the provided text and the architecture choices are presented as engineering decisions rather than theorems that presuppose the result.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The work implicitly assumes standard DNN training converges to a useful joint detector and that simulated channels match deployment conditions.

pith-pipeline@v0.9.0 · 5768 in / 1126 out tokens · 24060 ms · 2026-05-25T09:29:10.341052+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages

[1]

Space modulation on wireless fading channels,

Y . A. Chau and Shi-Hong Y u, “Space modulation on wireless fading channels,” in IEEE 54th V ehicular Technology Conference. VTC Fall

work page
[2]

No.01CH37211) , vol

Proceedings (Cat. No.01CH37211) , vol. 3, Oct 2001, pp. 1668– 1671 vol.3

work page 2001
[3]

Spatial modulation,

R. Y . Mesleh, H. Haas, S. Sinanovic, C. W. Ahn, and S. Y un, “ Spatial modulation,” IEEE IEEE Trans. V eh. Technol., vol. 57, no. 4, pp. 2228– 2241, July. 2008

work page 2008
[4]

The challenges facing physical layer securi ty,

W. Trappe, “The challenges facing physical layer securi ty,” IEEE Com- mun. Mag. , vol. 53, no. 6, pp. 16–20, June 2015

work page 2015
[5]

Two high-perf ormance schemes of transmit antenna selection for secure spatial mo dulation,

F. Shu, Z. Wang, R. Chen, Y . Wu, and J. Wang, “Two high-perf ormance schemes of transmit antenna selection for secure spatial mo dulation,” IEEE Trans. V eh. Technol., vol. 67, no. 9, pp. 8969–8973, Sep. 2018

work page 2018
[6]

Secure and precise wireless transmission for random-subcarrier-selection- based directional modulation transmit antenna array,

F. Shu, X. Wu, J. Hu, J. Li, R. Chen, and J. Wang, “Secure and precise wireless transmission for random-subcarrier-selection- based directional modulation transmit antenna array,” IEEE J. Sel. Areas Commun. , vol. 36, no. 4, pp. 890–904, Apr. 2018

work page 2018
[7]

Antenna selection method of maximizing secrecy rate for green secur e spatial modulation,

G. Xia, F. Shu, Y . Zhang, J. Wang, S. ten Brink, and J. Speid el, “Antenna selection method of maximizing secrecy rate for green secur e spatial modulation,” IEEE Trans. Green Commun and Netw. , vol. 3, no. 2, pp. 288–301, Jun. 2019

work page 2019
[8]

Secret precoding-ai ded spatial modulation,

F. Wu, L. Y ang, W. Wang, and Z. Kong, “Secret precoding-ai ded spatial modulation,” IEEE Commun. Lett. , vol. 19, no. 9, pp. 1544–1547, Sep. 2015

work page 2015
[9]

Deep learning in neural networks: An ov erview,

J. Schmidhuber, “Deep learning in neural networks: An ov erview,” Neural Netw , vol. 61, pp. 85–117, 2015

work page 2015
[10]

M achine learning paradigms for next-generation wireless networks ,

C. Jiang, H. Zhang, Y . Ren, Z. Han, K. Chen, and L. Hanzo, “M achine learning paradigms for next-generation wireless networks ,” Wireless Commun., vol. 24, no. 2, pp. 98–105, Apr. 2017

work page 2017
[11]

Deep learning-based c hannel estimation for beamspace mmwave massive mimo systems,

H. He, C. Wen, S. Jin, and G. Y . Li, “Deep learning-based c hannel estimation for beamspace mmwave massive mimo systems,” IEEE Wireless Communications Letters, vol. 7, no. 5, pp. 852–855, Oct. 2018

work page 2018
[12]

Learning to d ecode linear codes using deep learning,

E. Nachmani, Y . Be’ery, and D. Burshtein, “Learning to d ecode linear codes using deep learning,” in 2016 54th Annual Allerton Conference on Communication, Control, and Computing (Allerton) , Sep. 2016, pp. 341–346

work page 2016
[13]

Deep lea rning for super-resolution channel estimation and DOA estimation ba sed massive MIMO system,

H. Huang, J. Y ang, H. Huang, Y . Song, and G. Gui, “Deep lea rning for super-resolution channel estimation and DOA estimation ba sed massive MIMO system,” IEEE Trans. V eh. Technol. , vol. 67, no. 9, pp. 8549– 8560, Aug. 2018

work page 2018
[14]

Deep learning-aided s cma,

M. Kim, N. Kim, W. Lee, and D. Cho, “Deep learning-aided s cma,” IEEE Commun. Lett. , vol. 22, no. 4, pp. 720–723, Apr. 2018

work page 2018
[15]

Deep learning methods for improved decoding o f linear codes,

E. Nachmani, E. Marciano, L. Lugosch, W. J. Gross, D. Bur shtein, and Y . Beery, “Deep learning methods for improved decoding o f linear codes,” IEEE J. Sel. Topics Signal Process. , vol. 12, no. 1, pp. 119–131, Feb. 2018

work page 2018
[16]

High-per formance power allocation strategies for secure spatial modulation ,

F. Shu, X. Liu, G. Xia, T. Xu, J. Li, and J. Wang, “High-per formance power allocation strategies for secure spatial modulation ,” IEEE Trans. V eh. Technol., vol. 68, no. 5, pp. 5164–5168, May 2019. FENG SHU is a professor with the School of Electronic and Opti cal Engineering at Nanjing University of Science and Technology, Nanjing, China. He is also wi...

work page 2019

[1] [1]

Space modulation on wireless fading channels,

Y . A. Chau and Shi-Hong Y u, “Space modulation on wireless fading channels,” in IEEE 54th V ehicular Technology Conference. VTC Fall

work page

[2] [2]

No.01CH37211) , vol

Proceedings (Cat. No.01CH37211) , vol. 3, Oct 2001, pp. 1668– 1671 vol.3

work page 2001

[3] [3]

Spatial modulation,

R. Y . Mesleh, H. Haas, S. Sinanovic, C. W. Ahn, and S. Y un, “ Spatial modulation,” IEEE IEEE Trans. V eh. Technol., vol. 57, no. 4, pp. 2228– 2241, July. 2008

work page 2008

[4] [4]

The challenges facing physical layer securi ty,

W. Trappe, “The challenges facing physical layer securi ty,” IEEE Com- mun. Mag. , vol. 53, no. 6, pp. 16–20, June 2015

work page 2015

[5] [5]

Two high-perf ormance schemes of transmit antenna selection for secure spatial mo dulation,

F. Shu, Z. Wang, R. Chen, Y . Wu, and J. Wang, “Two high-perf ormance schemes of transmit antenna selection for secure spatial mo dulation,” IEEE Trans. V eh. Technol., vol. 67, no. 9, pp. 8969–8973, Sep. 2018

work page 2018

[6] [6]

Secure and precise wireless transmission for random-subcarrier-selection- based directional modulation transmit antenna array,

F. Shu, X. Wu, J. Hu, J. Li, R. Chen, and J. Wang, “Secure and precise wireless transmission for random-subcarrier-selection- based directional modulation transmit antenna array,” IEEE J. Sel. Areas Commun. , vol. 36, no. 4, pp. 890–904, Apr. 2018

work page 2018

[7] [7]

Antenna selection method of maximizing secrecy rate for green secur e spatial modulation,

G. Xia, F. Shu, Y . Zhang, J. Wang, S. ten Brink, and J. Speid el, “Antenna selection method of maximizing secrecy rate for green secur e spatial modulation,” IEEE Trans. Green Commun and Netw. , vol. 3, no. 2, pp. 288–301, Jun. 2019

work page 2019

[8] [8]

Secret precoding-ai ded spatial modulation,

F. Wu, L. Y ang, W. Wang, and Z. Kong, “Secret precoding-ai ded spatial modulation,” IEEE Commun. Lett. , vol. 19, no. 9, pp. 1544–1547, Sep. 2015

work page 2015

[9] [9]

Deep learning in neural networks: An ov erview,

J. Schmidhuber, “Deep learning in neural networks: An ov erview,” Neural Netw , vol. 61, pp. 85–117, 2015

work page 2015

[10] [10]

M achine learning paradigms for next-generation wireless networks ,

C. Jiang, H. Zhang, Y . Ren, Z. Han, K. Chen, and L. Hanzo, “M achine learning paradigms for next-generation wireless networks ,” Wireless Commun., vol. 24, no. 2, pp. 98–105, Apr. 2017

work page 2017

[11] [11]

Deep learning-based c hannel estimation for beamspace mmwave massive mimo systems,

H. He, C. Wen, S. Jin, and G. Y . Li, “Deep learning-based c hannel estimation for beamspace mmwave massive mimo systems,” IEEE Wireless Communications Letters, vol. 7, no. 5, pp. 852–855, Oct. 2018

work page 2018

[12] [12]

Learning to d ecode linear codes using deep learning,

E. Nachmani, Y . Be’ery, and D. Burshtein, “Learning to d ecode linear codes using deep learning,” in 2016 54th Annual Allerton Conference on Communication, Control, and Computing (Allerton) , Sep. 2016, pp. 341–346

work page 2016

[13] [13]

Deep lea rning for super-resolution channel estimation and DOA estimation ba sed massive MIMO system,

H. Huang, J. Y ang, H. Huang, Y . Song, and G. Gui, “Deep lea rning for super-resolution channel estimation and DOA estimation ba sed massive MIMO system,” IEEE Trans. V eh. Technol. , vol. 67, no. 9, pp. 8549– 8560, Aug. 2018

work page 2018

[14] [14]

Deep learning-aided s cma,

M. Kim, N. Kim, W. Lee, and D. Cho, “Deep learning-aided s cma,” IEEE Commun. Lett. , vol. 22, no. 4, pp. 720–723, Apr. 2018

work page 2018

[15] [15]

Deep learning methods for improved decoding o f linear codes,

E. Nachmani, E. Marciano, L. Lugosch, W. J. Gross, D. Bur shtein, and Y . Beery, “Deep learning methods for improved decoding o f linear codes,” IEEE J. Sel. Topics Signal Process. , vol. 12, no. 1, pp. 119–131, Feb. 2018

work page 2018

[16] [16]

High-per formance power allocation strategies for secure spatial modulation ,

F. Shu, X. Liu, G. Xia, T. Xu, J. Li, and J. Wang, “High-per formance power allocation strategies for secure spatial modulation ,” IEEE Trans. V eh. Technol., vol. 68, no. 5, pp. 5164–5168, May 2019. FENG SHU is a professor with the School of Electronic and Opti cal Engineering at Nanjing University of Science and Technology, Nanjing, China. He is also wi...

work page 2019