DeepJSCC-f: Deep Joint Source-Channel Coding of Images with Feedback (1911.11174v2)

Abstract: We consider wireless transmission of images in the presence of channel output feedback. From a Shannon theoretic perspective feedback does not improve the asymptotic end-to-end performance, and separate source coding followed by capacity-achieving channel coding, which ignores the feedback signal, achieves the optimal performance. It is well known that separation is not optimal in the practical finite blocklength regime; however, there are no known practical joint source-channel coding (JSCC) schemes that can exploit the feedback signal and surpass the performance of separation-based schemes. Inspired by the recent success of deep learning methods for JSCC, we investigate how noiseless or noisy channel output feedback can be incorporated into the transmission system to improve the reconstruction quality at the receiver. We introduce an autoencoder-based JSCC scheme, which we call DeepJSCC-f, that exploits the channel output feedback, and provides considerable improvements in terms of the end-to-end reconstruction quality for fixed-length transmission, or in terms of the average delay for variable-length transmission. To the best of our knowledge, this is the first practical JSCC scheme that can fully exploit channel output feedback, demonstrating yet another setting in which modern machine learning techniques can enable the design of new and efficient communication methods that surpass the performance of traditional structured coding-based designs.

• The paper demonstrates that DeepJSCC-f achieves up to a 3 dB PSNR gain over traditional separation schemes in low SNR conditions.
• It employs layered autoencoders with iterative feedback integration to refine image quality during transmission.
• The approach is robust to noisy feedback and adapts transmission rates for efficient bandwidth utilization in variable channel conditions.

Overview of "DeepJSCC-f: Deep Joint Source-Channel Coding of Images with Feedback"

The paper "DeepJSCC-f: Deep Joint Source-Channel Coding of Images with Feedback," authored by David Burth Kurka and Deniz Gündüz, offers an innovative approach in the field of joint source-channel coding (JSCC) by leveraging deep learning models. The paper's pivotal contribution is the development of an autoencoder-based JSCC scheme, DeepJSCC-f, which strategically incorporates channel output feedback to enhance the wireless transmission quality of images. This paper challenges traditional separation-based schemes by demonstrating that the integration of feedback in JSCC can achieve superior performance in both fixed-length and variable-length transmission settings.

Technical Contributions

Central to the paper is the use of deep learning, specifically convolutional neural networks (CNNs), to establish a JSCC scheme that effectively exploits feedback. The main architecture consists of layered autoencoders, enabling the model to adjust transmissions based on feedback information, which translates into higher quality image reconstructions or reduced transmission delay. This novel approach allows the transmitter to refine the transmitted signal iteratively, using the received feedback after each transmission layer.

The paper makes strong claims, supported by empirical results, stating that DeepJSCC-f outperforms separation-based designs—those employing state-of-the-art image compression standards like JPEG, JPEG2000, or BPG, followed by capacity-achieving channel codes. The results clearly indicate the advantages of joint design, especially in environments with feedback, where the separation-based approach is strictly suboptimal due to its inability to adaptively leverage channel output information in real-time.

Key Experimental Findings

The numerical results presented in the paper are compelling. Notably, DeepJSCC-f achieves better performance than separation schemes, offering a gain of up to 3 dB in peak signal-to-noise ratio (PSNR) in low signal-to-noise ratio (SNR) conditions. Additionally, the paper demonstrates the proposed architecture's robustness against noisy feedback, a setting where traditional schemes falter. The model maintains performance under varying feedback channel conditions, illustrating its potential for practical adoption in diverse real-world scenarios.

DeepJSCC-f's layered structure also proves advantageous for variable-rate transmission. By leveraging feedback effectively, the system is able to dynamically adjust the number of transmission layers, thereby optimizing the required bandwidth while meeting specific quality targets.

Implications and Future Directions

The implications of this work are multifaceted. Theoretically, it extends the applicability of JSCC by incorporating feedback, challenging the long-held belief in the supremacy of separation under all realistic conditions. Practically, it provides a pathway for implementing more efficient communication systems, particularly in environments characterized by limited bandwidth and time-varying channel conditions.

Future research directions could explore further enhancements in the model architecture and training methodologies, as well as applications of the proposed JSCC scheme to other types of data, such as audio and video, complementing the techniques developed for images. Additionally, integrating adaptive power allocation strategies could further bolster the system's performance, particularly under fading channel conditions.

In conclusion, "DeepJSCC-f: Deep Joint Source-Channel Coding of Images with Feedback" marks a significant advancement in the intersection of deep learning and communication theory. It presents a practical approach that effectively exploits feedback, demonstrating the potential of deep neural networks to surpass traditional communication frameworks.