[Paper Review] Classification of Common Waveforms Including a Watchdog for Unknown Signals

C. Tanner Fredieu, Justin M. Bui+4 more|arXiv (Cornell University)|Aug 16, 2021

Wireless Signal Modulation Classification11 references4 citations

TL;DR

This paper proposes a two-stage deep learning framework for classifying common radar and communication waveforms (SC, SC-FDMA, OFDM, LFM) and detecting unknown signals in shared spectrum. It uses a deep multi-layer perceptron for classification and a CNN-based autoencoder with dual RMSE thresholds to flag unknown waveforms, achieving 100% classification accuracy above 0 dB SNR and 95% anomaly detection accuracy at 0 dB with optimized FFT size.

ABSTRACT

In this paper, we examine the use of a deep multi-layer perceptron model architecture to classify received signal samples as coming from one of four common waveforms, Single Carrier (SC), Single-Carrier Frequency Division Multiple Access (SC-FDMA), Orthogonal Frequency Division Multiplexing (OFDM), and Linear Frequency Modulation (LFM), used in communication and radar networks. Synchronization of the signals is not needed as we assume there is an unknown and uncompensated time and frequency offset. An autoencoder with a deep CNN architecture is also examined to create a new fifth classification category of an unknown waveform type. This is accomplished by calculating a minimum and maximum threshold values from the root mean square error (RMSE) of the radar and communication waveforms. The classifier and autoencoder work together to monitor a spectrum area to identify the common waveforms inside the area of operation along with detecting unknown waveforms. Results from testing showed the classifier had 100\% classification rate above 0 dB with accuracy of 83.2\% and 94.7\% at -10 dB and -5 dB, respectively, with signal impairments present. Results for the anomaly detector showed 85.3\% accuracy at 0 dB with 100\% at SNR greater than 0 dB with signal impairments present when using a high-value Fast Fourier Transform (FFT) size. Accurate detection rates decline as additional noise is introduced to the signals, with 78.1\% at -5 dB and 56.5\% at -10 dB. However, these low rates seen can be potentially mitigated by using even higher FFT sizes also shown in our results.

Motivation & Objective

• To address spectrum sharing challenges between radar and communication systems by enabling real-time waveform classification and unknown signal detection.
• To develop a robust, low-complexity classification model that operates under realistic signal impairments such as frequency offset, phase offset, and IQ imbalance.
• To design an anomaly detection system that identifies unknown waveforms using reconstruction error from a convolutional autoencoder.
• To improve detection accuracy by optimizing FFT size and using multi-threshold RMSE regions to reduce false positives from signal reconstruction gaps.

Proposed method

• A deep multi-layer perceptron (MLP) is trained to classify baseband signals into four categories: SC, SC-FDMA, OFDM, and LFM.
• Signals are converted to the frequency domain using an FFT, and the magnitude is taken before power spectral density (PSD) conversion for input to the anomaly detector.
• A convolutional autoencoder reconstructs input signals and computes root mean square error (RMSE) between input and output to detect anomalies.
• Two threshold values (minimum and maximum RMSE) are derived from known signal classes (radar and communication) to define a 'known' region; signals outside this range are flagged as unknown.
• Three-region and five-region threshold designs are tested to reduce the gap between radar and communication RMSE distributions, improving detection robustness.
• Higher FFT sizes (e.g., 16384) are used to stabilize performance under multi-path fading and AWGN, especially at low SNR.

Experimental results

Research questions

• RQ1 Can a deep feed-forward network achieve high-accuracy classification of common radar and communication waveforms under realistic impairments?
• RQ2 How effective is a CNN-based autoencoder with RMSE thresholding in detecting unknown waveforms in a shared spectrum environment?
• RQ3 What impact do different FFT sizes have on the stability and accuracy of anomaly detection under noise and fading?
• RQ4 Can multi-threshold region designs reduce false classification of unknown signals due to reconstruction error gaps?

Key findings

• The classifier achieved 100% accuracy above 0 dB SNR, with 94.7% accuracy at -5 dB and 83.2% at -10 dB under signal impairments.
• The anomaly detector achieved 95% accuracy at 0 dB SNR using a three-region threshold design, with performance declining to 78.1% at -5 dB and 56.5% at -10 dB under AWGN.
• Using a 16384-FFT size significantly improved detection stability and accuracy, especially under multi-path fading and low SNR conditions.
• The five-region threshold design outperformed two- and three-region designs by reducing the gap between radar and communication signal reconstruction errors, achieving >95% accuracy at 0 dB.
• Higher FFT sizes, particularly those close to the minimum signal length, enhanced resilience to AWGN and improved detection consistency.
• The system successfully classified known waveforms and detected unknown signals even when reconstruction errors of unknowns fell within the RMSE range of known signals, thanks to multi-threshold region design.