Uplink per UE average SINR

Hi,

it's option b) in linear domain over all transport blocks and PRBs used for transmission during the simulation. The individual SINR is post receiver processing (e.g. MRC combining).

In the uplink a 3GPP MMSE-FDE effective SINR model (reference is in the code) is used to average over the individual post-processing SINRs per PRB. I don't look at the individual symbols (REs) in the time domain and just generate one value per 1 ms TTI (subframe). If you were to look at each of the 14 symbols in time domain, you could have a normal effective SINR averaging over the post MMSE-FDE SINRs.

Cheers
Jan

Jan,

Thanks for the answer.

One further question about MMSE-FDM without inter-cell IRC

when designing the equalizer, I have few options

a) P*h / ( P*|h|^2 + N )

b) P*h / ( P*|h|^2 + N + \sum{ P_i } )

c) P*h / ( P*|h|^2 + N + \sum{ P_i |h_i|^2 } )

where N is the thermal noise level, P is received serving power, P_i s are the received interference power

For the case MIMO MMSE, I have other options

d) P*h / ( P*|h|^2 + N + Inter-Layer interference )

e) ..

To sum up, I'm confused about how I should model the "interference part" without inter-cell IRC

w/o the whole interference, w/o the interference fast fading, and w/o inter-layer interference

May I know which are you using for "SIMO MRC, MMSE-FDE" and "MIMO MMSE, MMSE-FDE" respectively?

BR,

Chiu

Hi,

I'm not sure I understand your question. In the uplink we use the
MMSE-FDE model to "compress" the individual (per PRB) SINRs into a
single SINR. The individual SINRs can be generated assuming a SIMO MRC
or MIMO MMSE and MIMO MMSE-IRC receiver model (actually, in the uplink
the uplink schedulers currently don't support tranmsission on multiple
UE antennas). And - if that is your question - yes, we consider the
instantaneous fading MIMO channels on the serving and all interfering
links when we compute the post receiver SINRs.

For further details, I suggest to look at the IMTAphy source code.

Best regards
Jan

2012/9/3 Chiu <email address hidden>:
> Question #207234 on IMTAphy changed:
> https://answers.launchpad.net/imtaphy/+question/207234
>
> Chiu posted a new comment:
> Jan,
>
> Thanks for the answer.
>
> One further question about MMSE-FDM without inter-cell IRC
>
> when designing the equalizer, I have few options
>
> a) P*h / ( P*|h|^2 + N )
>
> b) P*h / ( P*|h|^2 + N + \sum{ P_i } )
>
> c) P*h / ( P*|h|^2 + N + \sum{ P_i |h_i|^2 } )
>
> where N is the thermal noise level, P is received serving power, P_i s
> are the received interference power
>
>
> For the case MIMO MMSE, I have other options
>
> d) P*h / ( P*|h|^2 + N + Inter-Layer interference )
>
> e) ..
>
>
> To sum up, I'm confused about how I should model the "interference part" without inter-cell IRC
>
> w/o the whole interference, w/o the interference fast fading, and w/o
> inter-layer interference
>
> May I know which are you using for "SIMO MRC, MMSE-FDE" and "MIMO MMSE,
> MMSE-FDE" respectively?
>
> BR,
>
> Chiu
>
> --
> You received this question notification because you are an answer
> contact for IMTAphy.

Jan,

Let me explain my question more clearly.

First of all, I need to give my understanding of calculating effective SINR.

For downlink, we use OFDM and have different SINR values on different subcarriers and OFDM symbols.

     Typically, we aggregate/compress the different SINR values within a TB using EESM/MISEM to obtain the effective SINR.

For uplink, we use SC-FDMA. So we must calculate the SINR in time domain. In such case all the modulated symbol will actually see same value of SINR within one OFDM symbol, but may differ between different OFDM symbols.

I'm not sure if it is still applicable to use EESM/MIESM to compress the SINR values over the 14 OFDM symbols.
But off-course if you generate fast fading channel only per PRB, then a single SINR values can be calculated via the above method without any L2S SINR aggregation.

You mentioned that you use some MMSE-FDE SINR model. I'm not sure which one are you referring to. Is that the same as my method? Where can I find some reference documents?

As for my original question, I'm asking the design of MMSE equalizer, not the SINR calculation after MMSE equalizer.

In designing the MMSE coefficients, we need a SINR value. I'm asking for this SINR value, should we include ICI, fast fading of ICI, or purely thermal noise level.

After designing this MMSE coefficients, there comes the true post-SINR calculation based on previously mentioned method . And so this post-SINR calculation will surely depend on the MMSE coefficients, which in turn will be related to the SINR value in the MMSE equalizer design.

And for your codes, I'm actually looking at them. But the code is too advanced for me that I even can't find the main structure of your code. Do you have any suggestion on how to get start with your code? where is the main loop that I can trace?

Or it will be the best that if you have some code version that I can directly compile and debug using visual C++ :p

BTW, just curious whether do you attend the 3GPP LTE standard meeting currently?

BRs,

Chiu

Dear Chiu,

Visual C++ is not supported, but KDevelop is. You can run bin/setupKdevelop.py then run kdevelop and open openwns.kdev4. But I must admit Microsoft VS is much more convenient and stable than KDevelop. He had to stop Windows support due to missing Python libraries on Windows.

Most of the important code you should look at is in modules/phy/imtaphy/src/receivers especially "LinearReceiver.cpp".

Also look at modules/phy/imtaphy/src/link2system where you will find the "MMSE-FDE.hpp" class

Please look at 3GPP documents R1-050718 and R1-051352 for references on uplink SC-FDMA MMSE equalization model.

Note that power control is performed per TB (more precisely per layer) in LTE, so the only effect that could cause different received powers per subcarrier and symbol is small scale fading. For system level simulation small-scale fading is assumed constant per subchannel and TTI duration, so the simulator cannot provide values with higher resolution and therefore does not need to equalize / average etc.

Per PRB SINR is always calculated as transmitter side precoding, channel attenuation (channel coefficient) and receiver side filtering to all (signal, noise, interference). See LinearReceiver.cpp for details. Per PRB SINR values are then used to calculate Effective SINRs.

Dear Maciej,

So many thanks for the advices and the two documents. They are really useful!

In R1-050718, it gives HARQ IR model by directly approximating the parameter beta to be unchanged.

It appears a bit strange to me. Since in such model, can we assure the calculated effective SNR after re-transmission will always be larger than that before re-transmission ( which would certainly be true in the case of CC model ) ?

I'm saying this possibility because the EESM basically do sort of averaging. So declination of eff SNR by this IR model would happen if the SINR on the second transmission < SINR on the first transmission

If sadly effective SNR after re-transmission degrades, I would consider this IR model to be very inaccurate.

But in their simulation, it seems to be quite accurate, did I misunderstand anything?

In R1-051352, it indeed gives a SINR model for SC-FDMA. My original question about MMSE equalizer comes here.

See, on the whole derivation up to equation (13) in R1-051352, they assume no any inter cell interference.

There are two folds

1. Equalization weight W

       If we assume MMSE-IRC, then the gamma would be P_s / ( \sum {g_i P_i} + sigma_n )

       But if we assume MMSE without IRC, I don't know which of the following should we use

       a) P_s / ( \sum{P_i} + sigma_n )

       b) P_s / ( sigma_n )

2. SNR calculation

      The SNR formula in equation (13) is actually a special result that only true if we use the given MMSE equalizer in the scenario without inter cell interference.

But if we, say, use b) in the scenario with ICI, equation (13) would no longer be true. But still we can calculate the SINR value via a similar form of the first line in equation (12).

So to make sure:

In SLS, by using equation (13) and substituting the SNR as SINR per subcarrier/RB as an approximation (since this is not the correct calculation of true SINR) would be good enough? and you are using this method?

BRs,

Chiu

Dear Chiu,

I took the references from the code. I actually never looked in the documents. From my discussions with Jan I can tell you that the documents are only used to motivate and reference the different approach used for effective SINR calculation in the uplink.

Also a strong reason for all calibration results I know were created using CC, not IR, is a missing community agreed model for IR applicable for system level simulation. The CC is from IEEE 802.16m (WiMAX) evaluation methodology which was also used and referenced for LTE evaluation.

Greats,

Maciej

Hi Chiu and Maciej (thanks for answering),

Regarding SINR computation and MMSE filter computation:
Of course, you can find the details in the source code and Maciej
already hinted at where to look for it. A description of this is also
given in Chapter 13 of this upcoming book:
http://eu.wiley.com/WileyCDA/WileyTitle/productCd-1119976707.html

For the MMSE filter computation several models exist which basically
differ w.r.t. the available information concerning the effective serving
and interfering channels. We have the complete fast-fading MIMO channel
matrices available in the simulator but depending on how we model
the knowledge and estimation process of an actual receiver we use different
interference covariance matrix structures from "diagonal" to "exact". You
can find an overview on this topic in 3GPP TR 36.829 where a "Wishart"
covariance matrix estimation error model is introduced which we also
implemented.

One comment regarding CC/IR: The L2S model should be able to support IR
(i.e., for some parts you boost the SINR due to repetition just like in
CC and overall you change the code rate). The reason it is not currently
implemented is more on the transmitter side because you would have to
choose which redundancy version to transmit (added complexity) and also
figure out how exactly this changes the code rate and how many bits
are repeated.

Cheers
Jan

Dear Jan and Maciej ,

Thank you both for the answers.

Jan, thanks for pointing out the document TR 36.829, it clearly solve my confusion about MMSE and MMSE-IRC. And I'm really looking forward for the book.

As I understood in your words, you used either MMSE or MMSE-IRC (ideal or model the estimation error by Wishart distribution) over antenna domain to obtain post SINR per PRB (or per subcarrier, depending on the granularity in frequency domain). Then applying the SNR formula of equation (13) in R1-051352 on those SINRs to obtain the single value of SNR.

I instead derive the uplink SINR via similar procedure in R1-051352 with the same MMSE equalizer (without IRC) but with the existence of ICI when deriving the SINR. As I mentioned, the result formula would be slightly different from a direct application of equation (13) to the post SINRs.

Just to make sure that if you indeed use the former method I depict. I'm considering to also use that method to perhaps simplify my codes. So it that the case?

For HARQ modeling, I now understand that the change of the code rate would be modeled in the interpolation when applying the maps of SNR to BLER as you explained in previous thread. Even with EESM and approximating it by unchanged beta (hence so have a similar eff SNR after re-transmission), the BLER would be reduced by a different mapping. But to make sure again:

      Assume we have the SNR-BLER maps of code rate = 0.5 and 0.1. On the first transmission, we use code rate 0.5. Then we use a complete IR and have an effective code rate = 0.25 after the second transmission. Then should we find the BLER by this kind of formula? : find BLER such that SNR_eff = (15/40)*SNRmap_1(BLER) + (25/40)*SNRmap_2(BLER), where SNRmap_1 and SNRmap_2 are the maps of code rate 0.5 and 0.1 respectively.

      And the average of SNR should be taken in dB domain ?
      Though this is heuristic, why can we do such interpolation?

BRs,

Chiu

Thanks Jan, that solved my question.