Sequential Change Point Detection for PSK Modulation Switching
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Computer Science. The day in computer science: software, systems and theory on arXiv.
Tom: We're looking at the paper: "Sequential Change Point Detection for PSK Modulation Switching".
Jane: Here's the gist of the paper. Sequential Change Point Detection for PSK Modulation Switching studies the problem of sequential change-point detection (CPD) in received data when modulation schemes switch from one phase shift keying (PSK) to another over an additive white Gaussian noise channel Such scenarios arise in wireless communication systems when a transmitter deploys adaptive modulation and coding (AMC) and dynamically switches between different PSKs.
Tom: Towards this, the authors use two classical CPD algorithms; cumulative-sum (CUSUM) and moving-sum algorithm (MOSUM), implemented using the log-likelihood ratio (LLR) and its approximation.
Jane: Problem Formulation The work formulates the problem of detecting a transition from M′ -PSK to M-PSK as a sequential CPD problem and constructs CUSUM and MOSUM detectors using exact and approximate LLR statistics.
Tom: The system model considers a transmitter switching from M′ -PSK to M-PSK at an unknown time instant τ over an AWGN channel.
Paper summary: Tom: So to wrap up this discussion on "Sequential Change Point Detection for PSK Modulation Switching," what does this paper actually mean when you step back from the math?
Jane: It means they’ve taken a problem that shows up in adaptive modulation systems—where the transmitter switches between different PSK schemes—and provided concrete tools to find that exact moment of transition using sequential methods.
Lu: The authors constructed CUSUM and MOSUM detectors specifically by using the log likelihood ratio statistics, and they showed how their approximations relate back to performance metrics like ARL2FA and WADD.
Meng: For someone like me who has to deploy models at scale, the implication is that if we can reliably estimate these change points without running massive calculations every time, this method could significantly reduce the processing cost of our AMC systems.
Tom: I agree with Meng; the focus on comparing exact versus approximate methods and showing that the approximation holds most of the performance gives engineers confidence to use them instead of sticking strictly to the computationally heavy exact LLR calculations.
Jane: The paper’s conclusion is pretty straightforward: they show that their approximate CUSUM performs close enough to the exact version, and their analytical bounds accurately predict how the MOSUM method behaves in practice.
Lu: What's missing, though, as they flag in their own work is extending this analysis to transitions between arbitrary M prime PSK and M PSK orders. That’s an avenue for future research they point toward.
Tom: Yeah, so the next step is taking this sequential CPD approach and making it robust enough to handle any kind of modulation switch, not just the specific ones they focused on in this paper.
Jane: So, in simple terms, this paper gives us a practical toolkit—CUSUM and MOSUM—to detect when a wireless system flips its modulation scheme by looking at the sequence of log likelihood ratios generated over time.
Conclusion: Tom: So, this paper is about finding when a wireless transmitter switches from one phase shift keying signal to another over noise, page by page in a sequence of data.
Jane: Exactly what it is trying to figure out—that exact moment of switching and how we can spot it using sequential math.
Lu: The authors use two classical methods, CUSUM and MOSUM, built around these log likelihood ratio statistics to pinpoint that change point tau.
Meng: From my side at the startup, what I'm really looking at is whether this detection method is fast enough to run on live traffic without killing our latency.
Lalam: The paper shows that for the CUSUM method, Wald’s approximation gives us a way to estimate the average time it takes to detect a change point.
Tom: And for MOSUM, they use results from Bohm and Hackl to set bounds on how long it might take to find that switch, which is crucial because we don't always have infinite time.
Jane: They did some simulations comparing exact versus approximate versions of these methods and found the approximate CUSUM was very close to the exact one in terms of detection delay.
Meng: That’s good news for deployment because it means we can likely use a slightly simpler calculation that still gives us great performance without needing super powerful computers.
Lu: The authors suggest that these analytical bounds actually match how the MOSUM method performs when you run tests on real data, which is pretty strong evidence for the approach.
Tom: So, basically, they've taken complex signal processing and given us two solid tools to handle dynamic modulation switching in a wireless system.
Jane: It really boils down to giving engineers a concrete way to estimate those transition times when you're dealing with adaptive modulation and coding.
Lu: The real next step is seeing if we can take this idea and apply it to situations where the switch isn't just between two fixed PSK types, but between any M prime PSK and any M PSK.
Meng: I’m wondering how much of a real-world system would actually use this kind of detection pipeline in production right now.
IIIT Hyderabad
cs.IT, eess.SP, math.IT
Submitted: 2026-10-09
Updated: 2026-11-05
Comments: 12 Pages, 5 Figures, Online change-point detection for modulation transitions using CUSUM and MOSUM detectors
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 51/100
The gist: Sequential Change Point Detection for PSK Modulation Switching studies the problem of sequential change-point detection (CPD) in received data when modulation schemes switch from one phase shift…
Key concepts
- Sequential Change Point Detection (CPD)
- This is a technique used to find the exact moment in a continuous stream of data where something significant has changed. In this paper, it's applied to detect when a communication system switches from one modulation type (like M'-PSK) to another (like M-PSK).
- Log-Likelihood Ratio (LLR)
- The LLR is a statistical measure used to decide which of two possible hypotheses—in this case, whether the signal was sent using M-PSK or M'-PSK—is more likely based on the received data. It helps quantify the evidence for a modulation switch.
- CUSUM (Cumulative-Sum) Detector
- A CUSUM detector is a sequential algorithm that tracks the accumulation of LLRs over time. It is designed to be very sensitive to small, sustained shifts in the underlying data distribution, making it effective for pinpointing the exact timing of a change point.
- MOSUM (Moving-Sum) Detector
- The MOSUM detector is another sequential method that works by looking at a sliding window of recent LLRs. It accumulates the most recent LLR values within this window to detect abrupt changes in the signal characteristics.
Terminology
Summary
Sequential Change Point Detection for PSK Modulation Switching studies the problem of sequential change-point detection (CPD) in received data when modulation schemes switch from one phase shift keying (PSK) to another over an additive white Gaussian noise channel Such scenarios arise in wireless communication systems when a transmitter deploys adaptive modulation and coding (AMC) and dynamically switches between different PSKs. p. 0 p. 0 p. 0 p. 0 Towards this, the authors use two classical CPD algorithms; cumulative-sum (CUSUM) and moving-sum algorithm (MOSUM), implemented using the log-likelihood ratio (LLR) and its approximation. p. 0
Problem Formulation
The work formulates the problem of detecting a transition from M′ -PSK to M-PSK as a sequential CPD problem and constructs CUSUM and MOSUM detectors using exact and approximate LLR statistics. The system model considers a transmitter switching from M′ -PSK to M-PSK at an unknown time instant τ over an AWGN channel. p. 0 The received signal is modeled based on the passband signal, and the goal is to estimate the location τ where the switch between modulation schemes happens. p. 1
Detection Methods
The proposed detection methods include:
-
CUSUM method for CPD, which sequentially accumulates LLRs lk using Eq. p. 2 (8). The CUSUM statistic is updated as Wlk = max[0, Wlk-1 + lk]. p. 2
-
MOSUM method for CPD, which considers a sliding window of size w ≥ 1 and accumulates the most recent w LLRs. p. 2 The MOSUM statistic is defined as Ylk = Pk k'=k-w+1lk'.
Performance Analysis
The performance of the sequential CPD procedure is characterized by the average run length to false alarm (ARL2FA A), the average detection delay (ADD), and the worst-case average detection delay (W ADD). For CUSUM, Wald’s approximation for ARL of the CUSUM procedure is given by Lemma 1. p. 4 For MOSUM, bounds on ARL2F A and W ADD are derived using the results provided by Bohm–Hackl. p. 5
Simulation Results
Simulations compare five methods: Exact CUSUM, Approx CUSUM, Kernel CUSUM, Exact MOSUM, and Approx MOSUM. The results show that Exact CUSUM achieves the lowest ADD across the plotted range. p. 7 The gap between the exact and approximate CUSUM methods remains small, indicating that the approximation retains most of the detection performance. p. 7
Conclusion
The work concludes by stating that the approximate CUSUM performs close to the exact CUSUM, and that analytical bounds accurately predict the empirical behavior of the MOSUM. Future work will focus on extending the CUSUM analysis to arbitrary M′ -PSK to M-PSK transitions. p. 8
APPENDIX A
PROOF OF THEOREM 1
The conditional MGF of the approximate LLR λk defined in Eq. (10) is given by Mλk(t θk) = 1/sqrt[1 - 4t 2] exp[2γst cos 2θk / (1 - 2t) - sin 2θk / (1 + 2t)] <ref:2610. p. 914208#pg6>.
APPENDIX B
PROOF OF THEOREM 2
The conditional PDF of the approximate LLR λk defined in Eq. (10) is given by fλk(x θk) = e-γs P∞ n=0 P2n m=0 2n m π(2n)! p. 9 2(2n+1) (2γs) n × tan[mπ/4 - θk] / (1 + sin 2θk) n × x(2 m-mx)/(m-n) x squared <ref:2610. p. 914208#pg11>.
APPENDIX C
PROOF OF THEOREM 3
The covariance of Yλi and Yλj is given by [Σk]i,j = ((w - d)σ squared if d < w, 0 if d ≥ w).
REFERENCES
[1] A. Goldsmith, Wireless communications. p. 0 Cambridge press, 2005 <ref:2610. p. 014208#pg2>
[2] P. C. Sapiano and J. D. Martin, “Maximum likelihood PSK classifier,” in Proceedings of the IEEE Military Communications Conference (MILCOM ’96), vol. p. 0 3, McLean, VA, USA, 1996, pp. 1010–1014 <ref:2610. p. 014208#pg3>
[3] C.-Y. Huan and A. Polydoros, “Likelihood methods for MPSK modulation classification,” IEEE Transactions on Communications, vol. p. 0 43, no. 2/3/4, pp. 1493–1504, 1995 <ref:2610. p. 014208#pg7>
[4] A. Polydoros and K. Kim, “On the detection and classification of quadrature digital modulations in broad-band noise,” IEEE Transactions on Communications, vol. p. 0 38, no. 8, pp. 1199–1211, Aug. 1990 <ref:2610. p. 014208#pg5>
[5] A. Abdi, O. A. Dobre, R. Choudhry, Y. Bar-Ness, and W. Su, “Modulation classification in fading channels using antenna arrays,” in Proceedings of the IEEE Military Communications Conference (MILCOM 2004), vol. p. 0 1, Monterey, CA, USA, 2004 <ref:2610. p. 014208#pg6>
[6] P. Panagiotou, A. Anastasopoulos, and A. Polydoros, “Likelihood ratio tests for modulation classification,” in MILCOM 2000. p. 0 21st Century Military Communications Conference Proceedings, vol. 2, Los Angeles, CA, USA, 2000 <ref:2610. p. 0 p. 014208#pg7>
[7] A. Swami and B. M. Sadler, “Hierarchical digital modulation classification using cumulants,” IEEE Transactions on Communications, vol. p. 0 48, no. 3, pp. 416–429, Mar. 2000 <ref:2610.14208#pg8>
[8] O. A. Dobre, Y. Bar-Ness, and W. Su, “Higher-order cyclic cumulants for high order modulation classification,” in Proceedings of the IEEE Military Communications Conference (MILCOM 2003), vol. p. 0 1, 2003, pp. 112–117 <ref:2610. p. 014208#pg9>
[9] Z. Zhu and A. K. Nandi, Automatic Modulation Classification: Principles, Algorithms and Applications. p. 0 Wiley, 2015 <ref:2610. p. 014208#pg10>
[10] G. Vats, Arthi, D. Sequeira, and A. p. 0 Yardi, “Neural-network based change-point detection for binary linear block codes,” in 2026 IEEE Wireless Communications and Networking Conference (WCNC), 2026 <ref:2610. p. 014208#pg2>
[11] N. Waltman, “Sequential decision making for automatic modulation classification,” Ph. p. 0D. dissertation, Massachusetts Institute of Technology, 2019 <ref:2610. p. 014208#pg2>
[12] M. Basseville and I. Nikiforov, Detection of abrupt changes: theory and application. p. 0 prentice Hall Englewood Cliffs, 1993, vol. 104 <ref:2610. p. 014208#pg2>
[13] J. G. Proakis and D. G. Manolakis, Digital signal processing (3rd ed. p. 0): principles, algorithms, and applications. p. 0 USA: Prentice-Hall, Inc., 1996 <ref:2610. p. 014208#pg3>
[14] E. S. Page, “Continuous inspection schemes,” Biometrika, vol. p. 0 41, no. 1/2, pp. 100–115, 1954 <ref:2610. p. 014208#pg3>
[15] W. Bohm and P. p. 5 Hackl, “Improved bounds for the average run length of ¨ control charts based on finite weighted sums,” The Annals of Statistics, pp. p. 0 1895–1899, 1990 <ref:2610. p. 014208#pg8>
[16] T. Flynn and S. Yoo, “Change detection with the kernel cumulative sum algorithm,” arXiv preprint arXiv:1903. p. 001661, 2019 <ref:2610. p. 014208#pg3>
[17] R. V. Hogg, J. W. McKean, and A. T. Craig, Introduction to Mathematical Statistics, 8th ed. p. 0, pearson, 2019 <ref:2610. p. 014208#pg3>
[18] G. Cui, X. Yu, S. Iommelli, and L. Kong, “Exact distribution for the product of two correlated gaussian random variables,” IEEE Signal Processing Letters, vol. p. 9 23, no. 11, pp. 1662–1666, 2016 <ref:2610. p. 014208#pg6>
[19] F. W. Olver, D. W. Lozier, R. F. Boisvert, and C. W. Clark, NIST Handbook of Mathematical Functions. p. 6 New York, NY: Cambridge University Press, 2010 <ref:2610. p. 014208#pg12>
--- Page 2 ---
Modulation Scheme M′ or M-PSK s′1(t),..., s′τ−1(t), sτ (t), sτ+1(t),. p. 1.. M′ -PSK M-PSK AWGN Channel r1(t),. p. 0.., rτ (t), rτ+1(t), ··· CP D τˆ Fig. p. 1 1: System model for the sequential change-point detection (CPD) problem for modulation schemes follows. p. 1
--- Page 3 ---
In this work, we consider a communication system in which the transmitter switches from M′ -PSK to M-PSK at an unknown time instant τ <ref:2610.14208#pg3>. The received signal passes through an additive white Gaussian noise (AWGN) channel <ref:2610. p. 014208#pg3>. Given this received data, the goal is to estimate the location τ, where the switch between modulation schemes happen <ref:2610. p. 114208#pg3>.
--- Page 4 ---
For M-PSK, the information-bearing phase θk is given by θk ∈ [0, 2π/M): m = 0,..., M - 1 <ref:2610. p. 014208#pg3>. The expressions for the waveform of M′ -PSK can be written in a similar manner <ref:2610. p. 114208#pg3>. The received signal rk(t) is given by rk(t) = (s'k(t) + nk(t), if k.
--- Page 5 ---
Let l k M-PSK and l k M′ -PSK denote the log likelihood of receiving rk(t) under the condition M-PSK and M′ -PSK are used respectively <ref:2610.14208#pg3>. The log likelihood ratio (LLR) is defined as lk = log P[rk(t) M]P[rk(t) M′] <ref:2610. p. 014208#pg3>.
--- Page 6 ---
For our problem of change-point detection corresponding to the received data r1(t), r2(t),..., we make use of the sequence of LLRs l1, l2,. p. 2.., and corresponding approximate LLRs λ1, λ2,. p. 2.., and provide a method to estimate the location of the change-point τ in Section III <ref:2610. p. 214208#pg3>.
Sequential Change Point Detection for PSK Modulation Switching studies the problem of sequential change-point detection (CPD) in received data when modulation schemes switch from one phase shift keying (PSK) to another over an additive white Gaussian noise channel Such scenarios arise in wireless communication systems when a transmitter deploys adaptive modulation and coding (AMC) and dynamically switches between different PSKs.
The authors formulate the problem of detecting a transition from M′ -PSK to M-PSK as a sequential CPD problem and construct CUSUM and MOSUM detectors using exact and approximate LLR statistics. The system model considers a transmitter switching from M′ -PSK to M-PSK at an unknown time instant τ <ref:2610. p. 014208#pg3>.
The performance of the sequential CPD procedure is characterized by the average run length to false alarm (ARL2F A), the average detection delay (ADD), and the worst-case average detection delay (W ADD). For CUSUM, Wald’s approximation for ARL of the CUSUM procedure is given by Lemma 1 <ref:2610. p. 414208#pg3>. For MOSUM, bounds on ARL2F A and W ADD are derived using the results provided by Bohm–Hackl <ref:2610. p. 514208#pg8>.
Simulations compare five methods: Exact CUSUM, Approx CUSUM, Kernel CUSUM, Exact MOSUM, and Approx MOSUM. The results show that Exact CUSUM achieves the lowest ADD across the plotted range <ref:2610. p. 7 p. 714208#pg4>. The gap between the exact and approximate CUSUM methods remains small, indicating that the approximation retains most of the detection performance <ref:2610. p. 7 p. 714208#pg10>.
--- Page 9 ---
Pruning redundant reasoning tokens
The authors study the problem of sequential change-point detection (CPD) in received data when modulation schemes switch from one phase shift keying (PSK) to another over an additive white Gaussian noise channel. Such scenarios arise in wireless communication systems when a transmitter deploys adaptive modulation and coding (AMC) and dynamically switches between different PSKs.
Simulations compare five methods: Exact CUSUM, Approx CUSUM, Kernel CUSUM, Exact MOSUM, and Approx MOSUM <ref:2610.14208#pg4>. The results show that Exact CUSUM achieves the lowest ADD across the plotted range <ref:2610.14208#pg10>. The gap between the exact and approximate CUSUM methods remains small, indicating that the approximation retains most of the detection performance <ref:2610.14208#pg10>.
The authors study the problem of sequential change-point detection (CPD) in received data when modulation schemes switch from one phase shift keying (PSK) to another over an additive white Gaussian noise channel. Such scenarios arise in wireless communication systems when a transmitter deploys adaptive modulation and coding (AMC) and dynamically switches between different PSKs.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improved detection of modulation scheme switching in wireless communication systems by employing sequential change-point detection (CPD) methods like CUSUM and MOSUM using log-likelihood ratios (LLR). This allows the system to identify the exact time instant when a transmitter switches from one phase shift keying (PSK) to another over an additive white Gaussian noise channel.
-
Enhanced performance characterization of CPD algorithms by deriving closed-form expressions for the probability density function and moment generating function of approximate LLRs for specific transitions, such as QPSK to BPSK. This enables precise analysis of the average run length to false alarm (ARL2FA) and worst-case average detection delay (WADD).
-
Development of robust decision-making strategies by incorporating bias correction terms into the cumulative sum (CUSUM) procedure when using approximate LLRs, ensuring a negative drift before a change point. This prevents false detections by managing the expected value of the approximate LLR, as suggested by setting b = 1/2 ((2γs)M/2).
-
Application of multivariate Gaussian approximations to moving sum statistics for higher-order PSK transitions to characterize ARL2FA and WADD bounds. This allows for theoretical guarantees on performance even when closed-form expressions for approximate LLR moments are difficult to derive.
-
Implementation of simulation-based verification procedures comparing exact and approximate CPD methods across various signal-to-noise ratio (SNR) levels and modulation transitions. This confirms that the analytical bounds accurately predict empirical behavior, validating the accuracy of the proposed approximations for both CUSUM and MOSUM detectors.
Sources
Related papers
- The Gaussian Multiuser Channel Capacity Region
- The mathematical landscape of partial information decomposition: A comprehensive review of properties and measures
- Electromagnetic Signal and Information Theory: A Continuous-Aperture Array Perspective
- Analytically Characterized Optimal Power Control for Signal-Level-Integrated Sensing, Computing and Communication in Federated Learning
- Information Spectrum Methods for Multiple-Access Channels--Part I: epsilon-Capacity Regions in the Mixed Gaussian Case
- Seed Recycling: Explicit Codes for Mixed Synchronization Errors Matching the Leading Term of the Gilbert--Varshamov Bound