A Blind Pre-Processor for Modulation Classification Applications in Frequency-Selective Non-Gaussian Channels

Abstract

This paper presents a new preprocessing stage that allows for the reliable classification of digital amplitude-phase modulated signals in a practical scenario where: 1) the classifier has no knowledge of the timing (symbol transition epochs) of the received signal; 2) the noise added in the channel is non-Gaussian; and 3) the fading experienced by the signal is frequency selective. The proposed preprocessor, which is based on the Gibbs sampling algorithm, is used to acquire timing information and to estimate the channel state and noise distribution parameters blindly, i.e., without knowledge of the received symbol sequence and the modulation scheme used. With the obtained estimates, in a second processing stage, the signal is then classified by using an appropriate (likelihood-or feature-based) classification algorithm. To quantify the performance of the proposed preprocessor, the probability of correct classification obtained by using the preprocessor with different classification algorithms is presented. It is shown that, by using the proposed preprocessor, modulation classification algorithms can perform well compared with clairvoyant classifiers assumed to be symbol synchronous with the received signal and to have perfect knowledge of the channel state and noise distribution.

Modulation Classification for MIMO-OFDM Signals via Approximate Bayesian Inference

“…Moreover, the proposed Gibbs sampler presents a graceful degradation in the presence of a model mismatch caused by channel coding, e.g., a decrease in the success rate by 6% with a code rate of 1/3. Future works include devising a Gibbs sampling scheme that accounts for the effects of the timing and carrier frequency offsets for MIMO systems following, e.g., [21], [44], [49]. In addition, the development of Bayesian classification techniques that address non-Gaussian noise is also a topic for further investigation.…”

“…Therefore, the conditional distribution p(A = a|s, h, σ 2 , y) is equal to zero if the transmitted symbols s do not belong to a. As a result, the Gibbs sampler may fail to converge to the posterior distribution (see, e.g., [21]). In order to alleviate the problem outlined above, we propose to adopt a prior distribution encoded on a latent Dirichlet BN G 2 shown in Fig.…”

“…Remark 4: When the SNR is high, the convergence speed is severely limited by the close-to-zero probabilities in the conditional distribution (21). Specifically, as the Gibbs sampling proceeds with its iterations, the samples of σ 2 tend to be small, making the relationship between y[n, k] and s mt [n, k] almost deterministic.…”

“…In [20], a systematic Bayesian solution based on the latent Dirichlet Bayesian Network (BN) is proposed, which generalizes and improves upon the work in [19]. A preprocessor for modulation classification is developed in [21], whereby the timing offset is estimated using grid-based Gibbs sampling 1 . We note that, while most algorithms rely on the assumption that the channels are flat fading or additive white Gaussian noise channels, the approaches in [19]- [21] are well-suited also for frequency selective fading channels.…”

How to choose a neural network architecture? – A modulation classification example

“…This classification process has many applications in wireless communications, including in autonomous multi-mode and software-defined radios. Blind modulation classification, i.e., without estimating any parameters for the wireless channel or the underlying signal, has become quite popular especially with the advancement of a variety of classification techniques, especially machine learning techniques [12]. Note that in this paper, we focus on deep learning techniques as they can be used to classify a large set of modulation schemes as discussed in [4], as opposed to simple feature-based traditional modulation classifiers.…”