# How can I perform semi-blind non-orthogonal successive interference cancellation (SIC) / source-separation with SISO in 4G/5G Downlink?

Semi-blind non-orthogonal successive interference cancellation in the single antenna case with 4G/5G signals (applies generally though).

This is very similar to NOMA in 5G, except that I have no timing knowledge of the other parties. I also do not know the specific sequences being used, as it could be 1/3 possible PSS sequences and 1/168 possible SSS sequences for every PSS (504 total sequences).

Idea is: you have $$N$$ base stations, in this case 2. BS1 and BS2. BS1 will have a higher relative power to BS2 by around 40dB, but any value can apply. You can perfectly demodulate BS1. You have only 1 antenna, assume you have an infinite dynamic range. BS1 and BS2 have independent multipath Rayleigh fading channels, with moderate correlation. You are also moving, BS1 and BS2 are stationary and have extremely good modulation quality and stability. Assume your receiver also has extremely stable/ideal sampling clocks and references. Detect the existence of BS2 and be able to demodulate some very simple reference sequences that are used to derive the Physical Cell ID. These sequences have very long code lengths and are repeated every 5ms. Ignoring the demodulation part: how can we subtract the signal of BS1 so that all we are left with is BS2.

Can be modeled as: $$y=x_1*h_1+x_2*h_2+n,$$ with $$*$$ being convolution, and $$h_1$$ and $$h_2$$ the channel impulse responses.

Assume SNR is greater than 30dB. Noise is zero mean white gaussian.

$$h_1,h_2$$ vary rapidly and have a coherence time of roughly 66us or roughly 1/15ms. There will be a multipath doppler shift due to moving objects in the environment.

You are given $$y$$, $$x_1$$, and nothing else.

The Goal would be to estimate $$h_1$$ perfectly such that it does not contain any of $$x_2$$, $$h_2$$. Subtract it off, and you would be left only with the right side of the equation. This is the ideal.

You don't know the timing synchronization of BS2. You can make priors about how BS2 could be constructed, but that dives into the details of 4G/5G frame structure.

All OFDM modulations, 15KHz subcarrier spacing, 15Ks/s, 72 subcarriers with no DC carrier. 1.92Msps, 1.4MHz bandwidth. (Standard LTE 6 PRB.) Complex baseband signals.

Does anyone know a good direction to go in solving this problem? Potentially some existing academic resources that have a similar problem but in a different frame?

• This is classic SIC. For ideas, read Y. Shen, T. Luo and M. Z. Win, "Neighboring Cell Search Techniques for LTE Systems," 2010 IEEE International Conference on Communications, and J. Axnäs, Y. -. E. Wang, M. Kamuf and N. Andgart, "Successive interference cancellation techniques for LTE downlink," 2011 IEEE 22nd International Symposium on Personal, Indoor and Mobile Radio Communications. Mar 11, 2022 at 15:13
• I have known about the paper, but they do not go into too much detail on the estimation. dl.dropboxusercontent.com/s/lyjdrsnrn9ehc2d/… Is this simply what I would need to implement in Matlab to accomplish it given my setup? It seems a little too simple. Do you know any existing Matlab implementations of SIC for this type of use case that I can derive what is happening? (so I can learn/understand it, not just use it) I have been researching SIC/MMSE but am still trying to grasp the overall concept. Mar 11, 2022 at 15:31
• Read through their channel model again, it seems that they present the SIC solution for a flat fading channel rather than a multipath fading channel. I may be able to derive it myself, I'll have to do some algebra and see where I end up hah. Mar 11, 2022 at 15:53
• The authors of the first paper did present the result for the multipath channel model in Section III.B. You can read this lecture for the general MMSE estimation. I don't have and know any source code for SIC. You may want to send an email to the authors of the papers. Mar 11, 2022 at 16:12