GFSK Demodulation Using Learned Sequence Correlation

Gorday, Paul; Erdöl, Nurgün; Zhuang, Hanqi

doi:10.1109/ccwc47524.2020.9031267

Cited by 5 publications

(6 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Here we extend the work in [15] and examine three topologies for application to noncoherent demodulation. Each topology uses a complex-valued feature detection layer, which may be characterized as coherent or noncoherent, followed by one or more real-valued classification layers.…”

Section: B Contributionsmentioning

confidence: 99%

“…It allows us to use complex-valued signal processing in the first layer for more general and efficient feature extraction from complex-valued signals, but it also simplifies the neural network architecture and training and takes advantage of the growing knowledge base of real-valued techniques and activation functions for the remaining layers. Variations of this technique have been used in recent applications of neural networks to radar and communication systems [13], [14], [15]. In [13], the real and imaginary parts of a twodimensional radar image are separately filtered by different convolutional filters in the first layer followed by nonlinear combining of the results.…”

Section: A Complex Neural Networkmentioning

confidence: 99%

“…This is a departure from the inner product operation used in typical convolutional layers and is not considered in this work. In [15], the authors of this paper use complex (Hermitian) inner products in the first layer neurons together with complex magnitude activation to implement a noncoherent demodulator for Bluetooth LE.…”

Section: A Complex Neural Networkmentioning

confidence: 99%

“…Each topology uses a complex-valued feature detection layer, which may be characterized as coherent or noncoherent, followed by one or more real-valued classification layers. The feature detection layers correspond conceptually to the those in [13], [15], and a significant contribution of this work is to provide a detailed mathematical framework for understanding these and earlier approaches in terms of training behavior and learned feature representation. Backpropagation equations for the complex feature layers are efficiently derived using Wirtinger calculus.…”

Section: B Contributionsmentioning

confidence: 99%

“…Remarks: The weight update in (15) does not provide any phase rotation of the input instance. If a set of phaserotated versions of an input x 1 were presented during batch-mode training, the average weight update would be x 1 (δ 1 e jθ 1 + δ 2 e jθ 2 + • • • + δ M e jθ M ).…”

Section: Proof Of Theoremmentioning

confidence: 99%

See 4 more Smart Citations

Complex-Valued Neural Networks for Noncoherent Demodulation

Gorday

Erdöl

Zhuang

2020

IEEE Open J. Commun. Soc.

Self Cite

View full text Add to dashboard Cite

Noncoherent demodulation is an attractive choice for many wireless communication systems. It requires minimal protocol overhead for carrier synchronization, and it is robust to radio impairments commonly found in low-cost transceivers. Machine learning techniques, such as neural networks and deep learning, offer additional benefits for these systems. Practical communication systems often include nonlinearities, non-stationarity, and non-Gaussian noise, which complicate mathematical derivation of optimum demodulators. Learning approaches can optimize demodulator performance directly from simulated or measured radio data, which is often plentiful in the design and verification of today's integrated transceivers. This paper examines several candidate neural network topologies for use in noncoherent demodulation and provides a mathematical framework for their comparison. Each is based on a complexvalued feature detection layer, which may be characterized as coherent or noncoherent, followed by one or more real-valued classification layers. Backpropagation equations for the noncoherent feature layer include a synchronization term that facilitates training with noncoherent input data. The coherent layer does not synchronize training data, however a noncoherent demodulator can still be constructed by increasing the coherent layer capacity and adding a max pooling layer to marginalize the unknown signal phase. A frequency classification example highlights the differences between the topologies and confirms that optimum noncoherent demodulation can be learned in the presence of AWGN and random phase offsets. The topologies considered here are suitable for noncoherent demodulation of power-efficient modulations such as FSK and ASK, which are typical in today's short-range wireless communication systems. It is hoped that such topologies will lead to a future common architecture that can support the wide range of modulation formats in this space.

show abstract

Section: B Contributionsmentioning

confidence: 99%