Page 52 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 6 – Wireless communication systems in beyond 5G era

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 6

a pseudo‑Deepcode by replacing the [3] D. J. C. MacKay, “Good error correcting codes based
Deepcode modulator with a modulator of order = 4. on very sparse matrices”, IEEE Transactions on In‑
Results show that the DEF‑LSTM code has better formation Theory, vol. 45, no. 2, pp. 399‑431, Mar.
performance compared to the pseudo‑Deepcode as its 1999.
BLER is signi icantly lower in the whole range of SNR that
[4] C. Berrou, A. Glavieux and P. Thitimajshima, “Near
we evaluated. The DEF‑LSTM code BLER gain over
Shannon limit error‑correcting coding and decod‑
pseudo‑Deepcode is larger than one order of magnitude
ing: turbo‑codes,” International Conference on Com‑
for SNR=5 dB and 6 dB. Moreover, the DEF‑LSTM code
munications, ICC’93,Geneva, Switzerland, pp. 1064‑
outperforms NR LDPC (dashed black curve) by at least
70, May 1993.
three orders of magnitude BLER for SNR ≥ 4 dB.
[5] S. Lin, D. Costello, Error Control Coding, Prentice‑
5. CONCLUSION AND FURTHER WORK Hall, Englewood Cliffs, NJ: 1983, 2011.
[6] Y. Polyanskiy, H. V. Poor and S. Verdú, “Feedback in
A new deep neural network‑based error correction en‑
non‑asymptotic regime,” IEEE Transactions on Infor‑
coder architecture for channels with feedback has been
mation Theory,, vol. 57, no. 8, pp. 4903‑4925, Aug.
presented. The new architecture generates parity sym‑
2011.
bols based on feedbacks in longer time windows com‑
pared to prior architectures, thereby introducing long‑
[7] J. Schalkwijk, “A coding scheme for additive noise
range dependencies between parity symbols within each channels with feedback‑I: No bandwidth con‑
codeword.
straint,” IEEE Transactions on Information Theory,
It has been shown that the codes designed according to vol. 12, no. 2, pp. 172‑182, Aug. 1966.
the DEF architecture achieve lower error rates than any
other code designed for channels with feedback. As long‑ [8] M. Horstein, “Sequential transmission using noise‑
range dependencies between parity symbols are a nec‑ less feedback,” IEEE Transactions on Information
essary ingredient of all good error correction codes, it is Theory, vol. 9, no. 3, pp. 136‑143, July 1963.
expected that further performance improvements can be
[9] J. M. Ooi and G. W. Wornell, “Fast iterative coding
obtained by increasing the length of the feedback time
windows. techniques for feedback channels,” IEEE Transac‑
tions on Information Theory,, vol. 44, no. 7, pp. 2960‑
Moreover, by a suitable selection of the modulation or‑
2976, Nov. 1998.
der, we showed that these codes can adapt to the
forward channel quality, thereby providing the [10] Z. Ahmad, Z. Chance, D. J. Love and C. Wang, “Con‑
maximum spectral ef iciency that is attainable for the catenated coding using linear schemes for Gaussian
given forward channel quality. broadcast channels with noisy channel output feed‑
In this work, DEF codes have been designed and evaluated back,” IEEE Transactions on Communications, vol. 63,
for forward channels impaired by additive white Gaus‑ no. 11, pp. 4576‑4590, Nov. 2015.
sian noise and noiseless feedback, where the forward SNR
[11] K. Vakilinia, S. V. S. Sudarsan, V. S. Ranganathan,
has been assumed to be perfectly known at design time
D. Divsalar and R. D. Wesel, “Optimizing transmis‑
(NN training) and during LLS. Code design with imperfect
sion lengths for limited feedback with non‑binary
SNR knowledge and evaluations in more realistic scenar‑
LDPC examples,” IEEE Transaction on communica‑
ios, such as channels with fading and noisy feedback, are
tions, vol. 64, no. 6, pp. 2245‑2257, June 2016.
interesting subjects that will need to be addressed in or‑
der to make these codes applicable in real transmission [12] Huawei, HiSilicon, “Performance evaluation of LDPC
systems. However, these topics require further thorough codes for NR eMBB data,” R1‑1713740, 3GPP RAN1
investigation and therefore are left for future works. meeting #90, Prague, Czech Republic, August21–25,
2017.
REFERENCES
[1] H. Kim, Y . Jiang, S. Kannan, S. Oh, P . Viswanath,
“Deepcode: feedback codes via deep learning,” IEEE
Journal on Selected Areas in Information Theory, vol.
1, no. 1, pp. 194‑206, May 2020.
[2] E. Arikan, “Channel polarization: a method for con‑
structing capacity‑achieving codes for symmetric
binary‑input memoryless channels,” IEEE Transac‑
tions on Information Theory, vol. 55, no. 7, pp. 3051‑
3073, July 2009.

40 © International Telecommunication Union, 2021