IWR1443BOOST: reading raw data from DC1000EVM

I see your point. However, our support is limited to mmwave studio related only. Having said that, I still try to provide as much help as possible. Here's the table which may help you figure out the difference between DCA1000 and TSW1400:

1. FPGA sends data in the following Ethernet packet format.

Hi Michelle,

Tim's question about what has changed in the data format between the TSW1400 and the DCA1000 boards is still unanswered. The Radar Studio user's guide lists 2 formats for DCA1000, one interleaved and one non-interleaved. How does one know which format is applied? There is no further reference to interleaving in the whole user's guide.

Thank you,
François.

Francois,

The interleaved and non-interleaved depends on the type of radar device. For xWR12xx/xWR14xx, only interleaved mode is supported. For xWR16xx, only non-interleaved mode is supported. See 22.6 to 22.9 of mmwave studio user guide.

Let me know if you have any question.

Thanks and Regards,

Michelle

Hi Ahmed,

I'm playing with the IWR1642 and DCA1000evm. I believe the matlab code you are referring to won't work in this instance as the data format has changed. If you look at table 22.8 and 22.9 in the mmwave studio user guide I think you will find the right format to modify the above script.

Also I found that the user guide was unclear whether the data for the DCA1000evm was in 2s complement or not i.e. needing to subtract 2^15 from the raw ADC. In matlab I've found doing an fread with 'int16' and not doing the subtraction gives the correct response.

Hi John,

thanks for the details.  As I have seen in the documentation, the IWR1642 and AWR1642 should have the same structure (xWR1642 non-interleaved) so the script should work for both devices ! 

which script are you using for the IWR1642 ? 

thank you in advance,

Regards,

Ahmed

Hi Ahmed,

I think you will find the original matlab code is for the TSW1400. The DCA1000evm has a slightly different format (I think according to the mmwave studio user guide).

This is my Matlab code that I modified to work with the demo they setup (when running the script DataCaptureDemo_xWR.lua in the quick start guide). I give no guarantee that it is correct but for me at least gives the right kind of results (you can compare the raw adc data with that from this matlab script and the post processing plots they provide to begin to convince yourself that it does the same thing). Also if you setup for TDM chirping or different number of receivers, frames etc you need to change the chirp parameters line 8 to 10. NOTE CODE IS FOR xWR1642 OTHER DEVICES MAY HAVE A DIFFERENT FORMAT AND THE INDICES IN THE CODE BELOW WOULD NEED TO CHANGE (see the mmwave studio guide for the correct IQ and channel format).

clear all
close all

fname = 'adc_data.bin';

% Chirp/data parameters - ONLY SETUP FOR 1TX AND 4RX (default example
% given).
slope_Hz_per_sec = 29.9817e12;
sampling_rate_Sps = 5000e3;
n_adc_samples = 256;
nChirps = 128;
n_frames = 8;
n_adc_bits = 16;  %%%%%%%% Change this to 12/16 based on what is ADC bit set in device %%%%
nBytesToUint16 = 2;
nRx = 4;
nTx = 1;
isreal = 0;

% Define some useful factors.
lightSpeed_meters_per_sec = 3e8;
norm_factor = (sqrt(2)/(2^(n_adc_bits-1)-1));

% Iterate over number of frames.
cdat = zeros(n_adc_samples, nRx*nTx, nChirps);

fid = fopen(fname,'r');

for frameId = 1:n_frames
    
    % Open the file and seek the current location for this frame.
    fseek(fid, frameId*(2*n_adc_samples)*nRx*nTx*nChirps*nBytesToUint16,'bof');

    % Read data corresponding to the chirp for each of the 4 receivers.
    dat = fread(fid,(2*n_adc_samples)*nRx*nTx*nChirps,'int16','l');

    dat = reshape(dat,n_adc_samples*2,nRx*nTx,[]);

    % Table 22.8 of mmwave_studio_user_guide.pdf implies following indices 
    % for real (R) and complex (C) data.
    indicesR1 = 1:4:n_adc_samples*2;
    indicesR2 = 2:4:n_adc_samples*2;
    indicesC1 = 3:4:n_adc_samples*2;
    indicesC2 = 4:4:n_adc_samples*2; 

    % The real and complex data is intermingled. Now reoder it to
    % create a complex variable (cdat). Deal with the real indices of
    % the data first.
    cdat(1:2:n_adc_samples,:,:) = dat(indicesR1,:,:) + 1i*dat(indicesC1,:,:);
    cdat(2:2:n_adc_samples,:,:) = dat(indicesR2,:,:) + 1i*dat(indicesC2,:,:);

    % Normalise the data.
    cdat = cdat*norm_factor;

    %% Specify window that will be applied.
    hann_winR = hanning(n_adc_samples);
    hann_winD = hanning(nChirps);


    % Form range doppler map for one receiver.
    for rxIdx = 1:1
        Y = (fft(squeeze(cdat(:,1,:))));
        Y2 = fftshift(fft((Y.')).' , 2);
        Y2 = abs(Y2);
        %fft2_out{rxIdx} = Y2;
    end

    % Define the x axis (range) and then plot.
    figure(91)
    range = ((((0:n_adc_samples-1)/n_adc_samples)*sampling_rate_Sps)/slope_Hz_per_sec)*lightSpeed_meters_per_sec/2;
    velocity = linspace(-0.09*nChirps/2,0.09*nChirps/2, nChirps);
    clims = [-20 60];
    imagesc(velocity,range,20*log10(Y2),clims)
    %imagesc(20*log10(Y2),clims)

    %figure(92)
    %surf(velocity,range,20*log10(Y2))

    pause(0.5)


end

fclose(fid);

Hope it works for you!

thank you john, I tried it but it seems that the dat and cdat do not have the same shape, I got an error as follows: "nonconformant arguments (op1 is 256x4x128, op2 is 128x4x0)". my configuration is : 4 rx, 2tx, 256 ADC samples, 128 chirps per frame, 8 frames.

am I missing something ?

Hi,

I believe that will be Time Division Multiplexing (TDM) setup. i.e. TX1 transmits and you collect all the chirps from the receivers and then TX2 transmits. At least that is my understanding.

I suspect the matlab error comes from line 36 where it only has rx and should have rx*tx. For my case it didn't matter as I only had it setup for 1TX.  I've edited the above code in previous post. However, the above code does not take into account TDM. For that you would have to change the indices slightly and reorder the chirps. My advice would be to get the above working for a 1TX and 4 RX scenario. Then modify it.

Best regards

Hi John,
I tried it with 4Rx and 1 Tx config and it is working but still not sure that I got the same figures as the PostProc ones.
I think to be sure, I should see the time domain signal and compare it by the postproc generated one.
the cdat contain the complex data generated (256,4,128), how to proceed to figure out the time domain from it ? I think it is better to format it as follows (as the TSW1600 in the guide does): rows should contains all of the data from the nth receiver ==> 4 rows, nb_columns = nb_ADC_samples * nb_chirps ==> 256*128 , then it is easy to plot the I and Q signal from each receiver, right ? 

thanks in advance,
Ahmed

Francois,

Sure. we will consider that request when generating application notes.

Thanks!
Regards,
Michelle

Hello everyone,

thanks a lot for your contribution. Let me summarize that we still have a problem with the post-processing of mulitple RAW files with respect to the UDP header and missing packets. Then it still remains unclear how to import and parse data in MATLAB for IWR1443 that has 4 LVDS lines and whose data format differs from that of AWR1642. Good news is that MATLAB treats the 'int16' data as 2's complement by default.

Looking forward to obtaining an application note, a UDP correction utility, and a script-example that shows how to read multiple RAW files in MATLAB. Thank you.

Regards,

Tim

Hi Michelle,

very nice, thanks! In return I would suggest to emphasize a couple of things in your application note:

1. Optimal working positions of all the switches on the DCA1000 board.

2. Optimal duty cycle that prevents the radar chip from overheating and yet does not slow down the measurement too much. My experiments show that 30% is a sort of margin which lets measure continuously for more than a few minutes. If you set it lower than you get an error after a few tens of seconds. If you set it higher your measurement cycle might become too long.

Unfortunately your MATLAB code does not organize the data over Tx / Rx / frames / chirps / samples.

Regards,

Tim

Michelle,

using your Matlab code I can correctly read and parse raw data from one data file after the UDP correction, see Figures below.  Now could you give an example of applying Packet_Reorder_Zerofill.exe or ar1.Packet_Reorder_Zerofill to multiple Raw bin-files? Section 3.2 in User's Guide looks very general but still I interpret it as I can process the files one by one usings the same log-file. Should I prepare a lua-script with my filenames or run a command in the command line of mm-Wave Studio? Thank you!

Regards,

Tim

Hi,

thank you Michelle, you mean normalize it by: norm_factor = (sqrt(2)/(2^(numADCBits-1)-1)) ?

this normalization factor does not normalize data between [0 1], should I normalize it depending on the maximum and minimum signal values ?

thanks in advance,

Ahmed

Hi Michelle,

Actually, for some machine learning uses, I need to normalize the data that's why it should be in the range [0 1] to reduce complexity. I am not computing the fft, I'm just using the signal as it is (one chirp per frame and so on.. ).

Thanks,

Ahmed

Hi michelle,

thank you. Do you know if there is some work done using radar data for some machine learning purposes ?

thanks,

Regards,

Ahmed

Hi Michelle,

thanks.

Regards,

Ahmed