| 1. |
- Brito, Flavio, et al.
(author)
-
Compression of Activation Signals from Split Deep Neural Network
- 2022
-
In: 2022 IEEE Latin-American Conference on Communications, LATINCOM 2022. - 9781665482257
-
Conference paper (peer-reviewed)abstract
- The use of artificial neural networks for the purpose of image classification, together with the advancement in computational capabilities of edge devices, plays an important role in the new emerging 5G use case scenarios. However, one of the main challenges of applications involving the use of these networks in edge devices is still the limitation of computational resources. An alternative for saving resources and promoting privacy are the split learning (or split inference) techniques, in which a deep neural network is cut into two parts and executed in distinct devices. Most of these techniques rely on sending the activation signals (output of the cut layer) through the communication channel. This work proposes a new compression algorithm for decreasing the bit rate required for the transmission of the activation signals (or simply 'scores'). The presented results demonstrate that the transmission rate can be decreased without hurting the neural network accuracy.
|
|
| 2. |
- Fonseca, Maria Nilma, et al.
(author)
-
DCT-based compression scheme for OFDM baseband signals
- 2018
-
In: 2018 Wireless Days, WD 2018. - 9781538656327 ; 2018-April, s. 92-97
-
Conference paper (peer-reviewed)abstract
- The requirements on the data rates of communication systems increase roughly an order of a magnitude at each new generation. This increase can be especially noticeable for the new centralized radio access networks (C-RAN) scenarios, considered for meeting the requirements of the fifth generation (5G) wireless systems. The general C-RAN architecture normally assumes data transmission in the form of digitized baseband, i.e. I/Q samples. This paper proposes a relatively low-complexity compression scheme for LTE OFDM baseband samples based on transform coding, using the discrete cosine transform (DCT). The compression method is evaluated in both downlink and uplink. The results show that this process can be configured to achieve compression ratio according with the load, for example, in the range of 5:1 to 31:1, with error vector magnitudes (EVMs) lower than 2.8%.
|
|
| 3. |
- Ramalho, Leonardo, et al.
(author)
-
An LPC-based fronthaul compression scheme
- 2017
-
In: IEEE Communications Letters. - 1089-7798. ; 21:2, s. 318-321
-
Journal article (peer-reviewed)abstract
- Several new architectures are under investigation for cloud radio access networks, assuming distinct splits of functionality among the network elements. Consequently, the research on radio data compression for the fronthaul is based on assumptions that correspond to a wide variety of tradeoffs among data rate, signal distortion, latency, and computational cost. This letter describes a method for LTE downlink point-topoint signal compression based on linear prediction and Huffman coding, which is suitable for low cost encoding and decoding units with stringent restrictions on power consumption. The proposed method can work at various compression factors, such as 3.3:1 at an average EVM of 0.9%, or 4:1 at an average EVM of 2.1%.
|
|
