def processBeamsOutput(csv_path, raytracing_path, rep, max):
    print('Generating Beams ...')
    # file generated with ak_generateInSitePlusSumoList.py:
    # need to use both LOS and NLOS here, cannot use restricted list because script does a loop over all scenes
    insiteCSVFile = csv_path
    numEpisodes = max  # total number of episodes
    # parameters that are typically not changed
    inputPath = raytracing_path + '/rosslyn_mobile_60GHz_ts0.1s_V_Lidar_e'
    normalizedAntDistance = 0.5
    angleWithArrayNormal = 0  # use 0 when the angles are provided by InSite

    # ULA
    number_Tx_antennas = 32
    number_Rx_antennas = 8

    # initialize variables
    numOfValidChannels = 0
    numOfInvalidChannels = 0
    numLOS = 0
    numNLOS = 0
    count = 0
    '''use dictionary taking the episode, scene and Rx number of file with rows e.g.:
    0,0,0,flow11.0,Car,753.83094753535,649.05232524135,1.59,D:/insitedata/noOverlappingTx5m/run00000,LOS=0
    0,0,2,flow2.0,Car,753.8198286576,507.38595866735,1.59,D:/insitedata/noOverlappingTx5m/run00000,LOS=1
    0,0,3,flow2.1,Car,749.7071175056,566.1905128583,1.59,D:/insitedata/noOverlappingTx5m/run00000,LOS=1'''

    with open(insiteCSVFile, 'r') as f:
        insiteReader = csv.DictReader(f)
        insiteDictionary = {}
        numExamples = 0
        for row in insiteReader:
            isValid = row[
                'Val']  # V or I are the first element of the list thisLine
            if isValid == 'V':  # filter the valid channels
                numExamples += 1
                thisKey = str(row['EpisodeID']) + ',' + str(
                    row['SceneID']) + ',' + str(row['VehicleArrayID'])
                insiteDictionary[thisKey] = row
        lastEpisode = int(row['EpisodeID'])
    allOutputs = np.nan * np.ones(
        (numExamples, number_Rx_antennas * number_Tx_antennas), np.float32)

    for e in range(numEpisodes):
        # print("Episode # ", e)
        b = h5py.File(inputPath + str(e) + '.hdf5', 'r')
        allEpisodeData = b.get('allEpisodeData')
        numScenes = allEpisodeData.shape[0]
        numReceivers = allEpisodeData.shape[1]
        # store the position (x,y,z), 4 angles of strongest (first) ray and LOS or not
        receiverPositions = np.nan * np.ones(
            (numScenes, numReceivers, 8), np.float32)
        # store two integers converted to 1
        episodeOutputs = np.nan * np.ones(
            (numScenes, numReceivers, number_Rx_antennas, number_Tx_antennas),
            np.float32)

        if (e % 50 == 0):
            print(e / numEpisodes)
        for s in range(numScenes):
            for r in range(numReceivers):  # 1
                insiteData = allEpisodeData[s, r, :, :]
                # if insiteData corresponds to an invalid channel, all its values will be NaN.
                # We check for that below
                numNaNsInThisChannel = sum(np.isnan(insiteData.flatten()))
                if numNaNsInThisChannel == np.prod(insiteData.shape):
                    numOfInvalidChannels += 1
                    continue  # next Tx / Rx pair
                thisKey = str(e) + ',' + str(s) + ',' + str(r)

                try:
                    thisInSiteLine = list(
                        insiteDictionary[thisKey].items())  # recover from dic
                except KeyError:
                    print('Could not find in dictionary the key: ', thisKey)
                    print('Verify file', insiteCSVFile)
                    exit(-1)
                # tokens = thisInSiteLine.split(',')
                if numNaNsInThisChannel > 0:
                    numOfValidRays = int(
                        thisInSiteLine[8]
                        [1])  # number of rays is in 9-th position in CSV list
                    # I could simply use
                    # insiteData = insiteData[0:numOfValidRays]
                    # given the NaN are in the last rows, but to be safe given that did not check, I will go for a slower solution
                    insiteDataTemp = np.zeros(
                        (numOfValidRays, insiteData.shape[1]))
                    numMaxRays = insiteData.shape[0]
                    validRayCounter = 0
                    for itemp in range(numMaxRays):
                        if sum(np.isnan(insiteData[itemp].flatten())
                               ) == 1:  # if insite version 3.2, else use 0
                            insiteDataTemp[validRayCounter] = insiteData[itemp]
                            validRayCounter += 1
                    insiteData = insiteDataTemp  # replace by smaller array without NaN
                receiverPositions[s, r, 0:3] = np.array([
                    thisInSiteLine[5][1], thisInSiteLine[6][1],
                    thisInSiteLine[7][1]
                ])

                numOfValidChannels += 1
                gain_in_dB = insiteData[:, 0]
                timeOfArrival = insiteData[:, 1]
                # InSite provides angles in degrees. Convert to radians
                # This conversion is being done within the channel function
                AoD_el = insiteData[:, 2]
                AoD_az = insiteData[:, 3]
                AoA_el = insiteData[:, 4]
                AoA_az = insiteData[:, 5]
                RxAngle = insiteData[:, 8][0]
                RxAngle = RxAngle + 90.0
                if RxAngle > 360.0:
                    RxAngle = RxAngle - 360.0
                # Correct ULA with Rx orientation
                AoA_az = -RxAngle + AoA_az  # angle_new = - delta_axis + angle_wi;

                # first ray is the strongest, store its angles
                receiverPositions[s, r, 3] = AoD_el[0]
                receiverPositions[s, r, 4] = AoD_az[0]
                receiverPositions[s, r, 5] = AoA_el[0]
                receiverPositions[s, r, 6] = AoA_az[0]

                isLOSperRay = insiteData[:, 6]
                pathPhases = insiteData[:, 7]

                # in case any of the rays in LOS, then indicate that the output is 1
                isLOS = 0  # for the channel
                if np.sum(isLOSperRay) > 0:
                    isLOS = 1
                    numLOS += 1
                else:
                    numNLOS += 1
                receiverPositions[s, r, 7] = isLOS
                mimoChannel = getNarrowBandULAMIMOChannel(
                    AoD_az, AoA_az, gain_in_dB, number_Tx_antennas,
                    number_Rx_antennas, normalizedAntDistance,
                    angleWithArrayNormal)
                equivalentChannel = np.abs(
                    getDFTOperatedChannel(mimoChannel, number_Tx_antennas,
                                          number_Rx_antennas))
                for rpt in range(0, rep):
                    mimoChannel = getNarrowBandULAMIMOChannel(
                        AoD_az, AoA_az, gain_in_dB, number_Tx_antennas,
                        number_Rx_antennas, normalizedAntDistance,
                        angleWithArrayNormal)
                    equivalentChannel = equivalentChannel + np.abs(
                        getDFTOperatedChannel(mimoChannel, number_Tx_antennas,
                                              number_Rx_antennas))
                episodeOutputs[s, r] = np.abs(equivalentChannel) / rep
                allOutputs[count] = np.ndarray.flatten(episodeOutputs[s, r])
                count += 1
            # finished processing this episode
    return allOutputs
Ejemplo n.º 2
0
def main():
    #file generated with ak_generateInSitePlusSumoList.py:
    #need to use both LOS and NLOS here, cannot use restricted list because script does a loop over all scenes
    insiteCSVFile = 'D:/github/5gm-data/list2_only_valids.csv'
    numEpisodes = 2086  #119  # total number of episodes
    outputFolder = 'D:/github/5gm-data/outputnn/'

    #parameters that are typically not changed
    if os.name == 'nt':
        #116 episodes
        #inputPath = 'D:/github/5gm-data/insitedata/urban_canyon_v2i_5gmv1_rays_e'
        #119 episodes
        #inputPath = 'D:/ak/Works/2018-proj-beam-sense-ml-lidar/lidar_and_insite/e119/insitedata/urban_canyon_v2i_5gmv1_positionMatrix_e'
        inputPath = 'D:/github/5gm-data/insitedata/urban_canyon_v2i_5gmv1_rays_e'
    else:
        #inputPath = '/mnt/d/github/5gm-data/insitedata/urban_canyon_v2i_5gmv1_rays_e'
        #inputPath = '/mnt/d/ak/Works/2018-proj-beam-sense-ml-lidar/lidar_and_insite/e119/insitedata/urban_canyon_v2i_5gmv1_positionMatrix_e'
        inputPath = '/mnt/d/github/5gm-data/insitedata/urban_canyon_v2i_5gmv1_rays_e'
    normalizedAntDistance = 0.5
    angleWithArrayNormal = 0  # use 0 when the angles are provided by InSite

    useUPA = True
    if useUPA == True:
        if False:
            number_Tx_antennasX = 16
            number_Tx_antennasY = 2
            number_Rx_antennasX = 4
            number_Rx_antennasY = 2
        else:
            #to get statistics:
            txCodebookInputFileName = 'D:/gits/lasse/software/mimo-matlab/tx_upa_codebook_16x16_N832_valid.mat'
            rxCodebookInputFileName = 'D:/gits/lasse/software/mimo-matlab/rx_upa_codebook_4x4_N52_valid.mat'
            #txCodebookInputFileName = 'D:/gits/lasse/software/mimo-matlab/tx_upa_codebook_12x12_valid.mat'
            #rxCodebookInputFileName = 'D:/gits/lasse/software/mimo-matlab/rx_upa_codebook_12x12_valid.mat'
            Wt, number_Tx_antennasX, number_Tx_antennasY, codevectorsIndicesTx = readUPASteeringCodebooks(
                txCodebookInputFileName)
            Wr, number_Rx_antennasX, number_Rx_antennasY, codevectorsIndicesRx = readUPASteeringCodebooks(
                rxCodebookInputFileName)
            number_Tx_vectors = Wt.shape[1]
            number_Rx_vectors = Wr.shape[1]

            if False:  #make one antenna at receiver
                Wr = None
                number_Rx_antennasX = 1
                number_Rx_antennasY = 1
                number_Rx_vectors = 1

            print(number_Tx_antennasX, number_Tx_antennasY,
                  number_Rx_antennasX, number_Rx_antennasY, number_Tx_vectors,
                  number_Rx_vectors)
        number_Tx_antennas = number_Tx_antennasX * number_Tx_antennasY
        number_Rx_antennas = number_Rx_antennasX * number_Rx_antennasY
    else:
        number_Tx_antennas = 32
        number_Rx_antennas = 8

    if not os.path.exists(outputFolder):
        os.makedirs(outputFolder)

    # initialize variables
    numOfValidChannels = 0
    numOfInvalidChannels = 0
    numLOS = 0
    numNLOS = 0
    numOccurrencesTxIndices = np.zeros((number_Tx_vectors, ), dtype=np.int)
    numOccurrencesRxIndices = np.zeros((number_Rx_vectors, ), dtype=np.int)
    numOccurrencesBeamPairIndices = np.zeros(
        (np.maximum(number_Tx_vectors, number_Rx_vectors)**2, ), dtype=np.int)
    '''
    use dictionary taking the episode, scene and Rx number of file with rows e.g.:
    0,0,0,flow11.0,Car,753.83094753535,649.05232524135,1.59,D:/insitedata/noOverlappingTx5m/run00000,LOS=0
    0,0,2,flow2.0,Car,753.8198286576,507.38595866735,1.59,D:/insitedata/noOverlappingTx5m/run00000,LOS=1
    0,0,3,flow2.1,Car,749.7071175056,566.1905128583,1.59,D:/insitedata/noOverlappingTx5m/run00000,LOS=1
    '''
    with open(insiteCSVFile, 'r') as f:
        insiteReader = csv.reader(f)
        insiteDictionary = {}
        for row in insiteReader:
            #print(row)
            thisKey = str(row[0]) + ',' + str(row[1]) + ',' + str(row[2])
            insiteDictionary[thisKey] = row

    for e in range(numEpisodes):
        print("Episode # ", e)
        # if using owncloud files
        # b = np.load('d:/github/5gm-data/insitedata/urban_canyon_v2i_5gmv1_rays_e'+str(e+1)+'.npz')
        b = np.load(inputPath + str(e) + '.npz')
        # b = np.load('./insitedata/urban_canyon_v2i_5gmv1_rays_e'+str(n+1)+'.npz')
        allEpisodeData = b['allEpisodeData']
        numScenes = allEpisodeData.shape[0]
        numReceivers = allEpisodeData.shape[1]
        #store the position (x,y,z), 4 angles of strongest (first) ray and LOS or not
        receiverPositions = np.nan * np.ones(
            (numScenes, numReceivers, 8), np.float32)
        #store two integers converted to 1
        episodeOutputs = np.nan * np.ones(
            (numScenes, numReceivers, number_Rx_vectors, number_Tx_vectors),
            np.float32)
        for s in range(numScenes):  # 50
            for r in range(numReceivers):  # 10
                insiteData = allEpisodeData[s, r, :, :]
                #if insiteData corresponds to an invalid channel, all its values will be NaN.
                #We check for that below
                numNaNsInThisChannel = sum(np.isnan(insiteData.flatten()))
                if numNaNsInThisChannel == np.prod(insiteData.shape):
                    #print('aaa', sum(np.isnan(insiteData.flatten())))
                    numOfInvalidChannels += 1
                    continue  # next Tx / Rx pair

                thisKey = str(e) + ',' + str(s) + ',' + str(r)
                try:
                    thisInSiteLine = insiteDictionary[
                        thisKey]  #recover from dic
                except KeyError:
                    print('Could not find in dictionary the key: ', thisKey)
                    print('Verify file', insiteCSVFile)
                    exit(-1)
                #5, 6, and 7
                #tokens = thisInSiteLine.split(',')
                if numNaNsInThisChannel > 0:
                    numOfValidRays = int(
                        thisInSiteLine[8]
                    )  #number of rays is in 9-th position in CSV list
                    #I could simply use
                    #insiteData = insiteData[0:numOfValidRays]
                    #given the NaN are in the last rows, but to be safe given that did not check, I will go for a slower solution
                    insiteDataTemp = np.zeros(
                        (numOfValidRays, insiteData.shape[1]))
                    numMaxRays = insiteData.shape[0]
                    validRayCounter = 0
                    for itemp in range(numMaxRays):
                        if sum(np.isnan(insiteData[itemp].flatten())) == 0:
                            insiteDataTemp[validRayCounter] = insiteData[itemp]
                            validRayCounter += 1
                    insiteData = insiteDataTemp  #replace by smaller array without NaN

                receiverPositions[s, r, 0:3] = np.array(
                    [thisInSiteLine[5], thisInSiteLine[6], thisInSiteLine[7]])

                numOfValidChannels += 1
                gain_in_dB = insiteData[:, 0]
                timeOfArrival = insiteData[:, 1]
                # InSite provides angles in degrees. Convert to radians
                # This conversion is being done within the channel function
                if True:  # use angles in degrees (convert later)
                    AoD_el = insiteData[:, 2]
                    AoD_az = insiteData[:, 3]
                    AoA_el = insiteData[:, 4]
                    AoA_az = insiteData[:, 5]
                else:  # convert now
                    AoD_el = np.deg2rad(insiteData[:, 2])
                    AoD_az = np.deg2rad(insiteData[:, 3])
                    AoA_el = np.deg2rad(insiteData[:, 4])
                    AoA_az = np.deg2rad(insiteData[:, 5])

                #AK TODO check
                if False:
                    AoA_az = AoA_az - 90
                    AoD_az = AoD_az - 90

                #first ray is the strongest, store its angles
                receiverPositions[s, r, 3] = AoD_el[0]
                receiverPositions[s, r, 4] = AoD_az[0]
                receiverPositions[s, r, 5] = AoA_el[0]
                receiverPositions[s, r, 6] = AoA_az[0]

                isLOSperRay = insiteData[:, 6]
                pathPhases = insiteData[:, 7]

                if False:  # enable for debugging with fixed angles
                    ad = (np.pi /
                          4) * 180 / np.pi  # in degrees, as InSite provides
                    aa = (np.pi / 2) * 180 / np.pi
                    ed = (np.pi /
                          6) * 180 / np.pi  # in degrees, as InSite provides
                    ea = -(np.pi / 5) * 180 / np.pi
                    g = 10
                    AoD_az = ad * np.ones(AoD_az.shape, AoD_az.dtype)
                    AoA_az = aa * np.ones(AoA_az.shape, AoA_az.dtype)
                    AoD_el = ed * np.ones(AoD_az.shape, AoD_az.dtype)
                    AoA_el = ea * np.ones(AoA_az.shape, AoA_az.dtype)
                    gain_in_dB = g * np.ones(gain_in_dB.shape,
                                             gain_in_dB.dtype)
                    pathPhases = np.zeros(pathPhases.shape)

                # order the rays to have the shortest path first
                # [timeOfArrival,sortedIndices] = sort(timeOfArrival);
                # theseRays=[];
                # theseRays.gainMagnitude = gainMagnitude(sortedIndices);
                # theseRays.timeOfArrival = timeOfArrival;
                # theseRays.AoA_el = AoA_el(sortedIndices); %not currently used
                # theseRays.AoD_el = AoD_el(sortedIndices); %not currently used
                # theseRays.AoA_az = AoA_az(sortedIndices);
                # theseRays.AoD_az = AoD_az(sortedIndices);
                # theseRays.isLOS = isLOS(sortedIndices);

                # in case any of the rays in LOS, then indicate that the output is 1
                isLOS = 0  # for the channel
                if np.sum(isLOSperRay) > 0:
                    isLOS = 1
                    numLOS += 1
                else:
                    numNLOS += 1
                receiverPositions[s, r, 7] = isLOS
                if useUPA == True:
                    #departure_angles = np.array((AoD_el,AoD_az)).T
                    #arrival_angles = np.array((AoA_el,AoA_az)).T
                    #calc_rx_power(departure_angles, arrival_angles, gain_in_dB, number_Tx_antennas, frequency=6e10)

                    mimoChannel = getNarrowBandUPAMIMOChannel(
                        AoD_el, AoD_az, AoA_el, AoA_az, gain_in_dB, pathPhases,
                        number_Tx_antennasX, number_Tx_antennasY,
                        number_Rx_antennasX, number_Rx_antennasY,
                        normalizedAntDistance)
                    equivalentChannel = getCodebookOperatedChannel(
                        mimoChannel, Wt, Wr)
                else:
                    mimoChannel = getNarrowBandULAMIMOChannel(
                        AoD_az, AoA_az, gain_in_dB, number_Tx_antennas,
                        number_Rx_antennas, normalizedAntDistance,
                        angleWithArrayNormal)
                    equivalentChannel = getDFTOperatedChannel(
                        mimoChannel, number_Tx_antennas, number_Rx_antennas)
                equivalentChannelMagnitude = np.abs(equivalentChannel)
                #print('equivalentChannelMagnitude  = ', equivalentChannelMagnitude)
                bestBeamPairIndex = np.argmax(equivalentChannelMagnitude,
                                              axis=None)
                numOccurrencesBeamPairIndices[bestBeamPairIndex] += 1
                #now it's not a simple unravel. Need to undo upa_codebook_creation.m association and
                #the Kronecker operation
                (bestRxIndex, bestTxIndex) = np.unravel_index(
                    bestBeamPairIndex, equivalentChannelMagnitude.shape)
                (bestRxIndex_xaxis,
                 bestRxIndex_yaxis) = codevectorsIndicesRx[bestRxIndex]
                (bestTxIndex_xaxis,
                 bestTxIndex_yaxis) = codevectorsIndicesTx[bestTxIndex]

                if False:
                    print('LOS,rx,tx = ', isLOS, bestRxIndex, bestTxIndex)
                    print('bestRxIndex_xaxis = ', bestRxIndex_xaxis)
                    print('bestRxIndex_yaxis = ', bestRxIndex_yaxis)
                    print('bestTxIndex_xaxis = ', bestTxIndex_xaxis)
                    print('bestTxIndex_yaxis = ', bestTxIndex_yaxis)
                #if isLOS == 0:
                numOccurrencesTxIndices[bestTxIndex] += 1  # increment counters
                numOccurrencesRxIndices[bestRxIndex] += 1
                # if bestRxIndex + bestTxIndex != 0:
                # print('bestRxIndex: ', bestRxIndex, ' and bestTxIndex: ', bestTxIndex)
                #    exit(1)
                outputLabel = bestTxIndex * number_Rx_antennas + bestRxIndex
                # when one needs to recover the labels:
                #recoverLabelTx = np.floor(outputLabel/number_Rx_antennas)
                #recoverLabelRx = outputLabel - recoverLabelTx*number_Rx_antennas
                episodeOutputs[s, r] = np.abs(equivalentChannel)

                #check if there is NaN. This can be disabled for speed, it's just for debugging
                if np.sum(np.isnan(episodeOutputs[s, r][:])) > 0:
                    print('Found Nan (e,s,r) = ', e, s, r)
                    exit(-1)

            #finished processing this episode
        npz_name = outputFolder + 'output_e_' + str(e) + '.npz'
        np.savez(npz_name, output=episodeOutputs)
        print('Saved file ', npz_name)

        outputFileName = outputFolder + 'outputs_positions_e_' + str(
            e) + '.hdf5'
        f = h5py.File(outputFileName, 'w')
        f['episodeOutputs'] = episodeOutputs
        f['receiverPositions'] = receiverPositions
        f.close()
        print('==> Wrote file ' + outputFileName)

    print('total numOfInvalidChannels = ', numOfInvalidChannels)
    print('total numOfValidChannels = ', numOfValidChannels)
    print('Sum = ', numOfValidChannels + numOfInvalidChannels)

    print('total numNLOS = ', numNLOS)
    print('total numLOS = ', numLOS)
    print('Sum = ', numLOS + numNLOS)

    #print('Statistics for NLOS only:')
    print('tx_indices_histogram = [', end=" ")
    for i in range(len(numOccurrencesTxIndices)):
        print(numOccurrencesTxIndices[i], end=" ")
    print('];')
    print('rx_indices_histogram = [', end=" ")
    for i in range(len(numOccurrencesRxIndices)):
        print(numOccurrencesRxIndices[i], end=" ")
    print('];')
    if False:
        print('Maximum among beam pair indices histogram:')
        print(np.amax(numOccurrencesBeamPairIndices))
        print('Beam pair indices histogram:')
        for i in range(len(numOccurrencesBeamPairIndices)):
            print(numOccurrencesBeamPairIndices[i], ' ')
def dataset_generation(cfg_file):
    tree = ET.parse(cfg_file)
    root = tree.getroot()

    cfg = {}
    for item in root.iter("item"):
        # print(item.attrib['name'], item.text)
        if item.text.isdigit():
            cfg[item.attrib["name"]] = int(item.text)
        elif isfloat(item.text):
            cfg[item.attrib["name"]] = float(item.text)
        else:
            cfg[item.attrib["name"]] = item.text

    limit = processCoordinates(
        cfg["data_folder"],
        cfg["dataset"],
        [cfg["rsu_x"], cfg["rsu_y"], cfg["rsu_z"]],
        [cfg["area_x_l"], cfg["area_y_l"], cfg["area_x_r"], cfg["area_y_r"]],
    )

    if not os.path.exists(cfg["outputFolder"]):
        os.makedirs(cfg["outputFolder"])

    # initialize variables
    numOfValidChannels = 0
    numOfInvalidChannels = 0
    numLOS = 0
    numNLOS = 0
    count = 0
    """use dictionary taking the episode, scene and Rx number of file with rows e.g.:
    0,0,0,flow11.0,Car,753.83094753535,649.05232524135,1.59,D:/insitedata/noOverlappingTx5m/run00000,LOS=0
    0,0,2,flow2.0,Car,753.8198286576,507.38595866735,1.59,D:/insitedata/noOverlappingTx5m/run00000,LOS=1
    0,0,3,flow2.1,Car,749.7071175056,566.1905128583,1.59,D:/insitedata/noOverlappingTx5m/run00000,LOS=1"""

    with open(cfg["insiteCSVFile"], "r") as f:
        insiteReader = csv.DictReader(f)
        insiteDictionary = {}
        numExamples = 0
        for row in insiteReader:
            isValid = row[
                "Val"]  # V or I are the first element of the list thisLine
            if isValid == "V":  # filter the valid channels
                numExamples += 1
                thisKey = (str(row["EpisodeID"]) + "," + str(row["SceneID"]) +
                           "," + str(row["VehicleArrayID"]))
                insiteDictionary[thisKey] = row
        lastEpisode = int(row["EpisodeID"])
    allOutputs = np.nan * np.ones(
        (numExamples, cfg["number_Rx_antennas"], cfg["number_Tx_antennas"]),
        np.complex128,
    )

    for e in range(cfg["numEpisodes"]):
        # print("Episode # ", e)
        b = h5py.File(cfg["inputPath"] + str(e) + ".hdf5", "r")
        allEpisodeData = b.get("allEpisodeData")
        numScenes = allEpisodeData.shape[0]
        numReceivers = allEpisodeData.shape[1]
        # store the position (x,y,z), 4 angles of strongest (first) ray and LOS or not
        receiverPositions = np.nan * np.ones(
            (numScenes, numReceivers, 8), np.float32)
        # store two integers converted to 1
        episodeOutputs = np.nan * np.ones(
            (
                numScenes,
                numReceivers,
                cfg["number_Rx_antennas"],
                cfg["number_Tx_antennas"],
            ),
            np.float32,
        )

        for s in range(numScenes):
            for r in range(numReceivers):  # 10
                insiteData = allEpisodeData[s, r, :, :]

                # if insiteData corresponds to an invalid channel, all its values will be NaN.
                # We check for that below
                numNaNsInThisChannel = sum(np.isnan(insiteData.flatten()))
                if numNaNsInThisChannel == np.prod(insiteData.shape):
                    numOfInvalidChannels += 1
                    continue  # next Tx / Rx pair
                thisKey = str(e) + "," + str(s) + "," + str(r)

                try:
                    thisInSiteLine = list(
                        insiteDictionary[thisKey].items())  # recover from dic
                except KeyError:
                    print("Could not find in dictionary the key: ", thisKey)
                    print("Verify file", insiteCSVFile)
                    exit(-1)
                # tokens = thisInSiteLine.split(',')
                if numNaNsInThisChannel > 0:
                    numOfValidRays = int(
                        thisInSiteLine[8]
                        [1])  # number of rays is in 9-th position in CSV list
                    # I could simply use
                    # insiteData = insiteData[0:numOfValidRays]
                    # given the NaN are in the last rows, but to be safe given that did not check, I will go for a slower solution
                    insiteDataTemp = np.zeros(
                        (numOfValidRays, insiteData.shape[1]))
                    numMaxRays = insiteData.shape[0]
                    validRayCounter = 0
                    for itemp in range(numMaxRays):
                        if (sum(np.isnan(insiteData[itemp].flatten())) == 1
                            ):  # if insite version 3.2, else use 0
                            insiteDataTemp[validRayCounter] = insiteData[itemp]
                            validRayCounter += 1
                    insiteData = insiteDataTemp  # replace by smaller array without NaN
                receiverPositions[s, r, 0:3] = np.array([
                    thisInSiteLine[5][1], thisInSiteLine[6][1],
                    thisInSiteLine[7][1]
                ])

                numOfValidChannels += 1
                gain_in_dB = insiteData[:, 0]
                timeOfArrival = insiteData[:, 1]
                # InSite provides angles in degrees. Convert to radians
                # This conversion is being done within the channel function
                AoD_el = insiteData[:, 2]
                AoD_az = insiteData[:, 3]
                AoA_el = insiteData[:, 4]
                AoA_az = insiteData[:, 5]
                RxAngle = insiteData[:, 8][0]
                RxAngle = RxAngle + 90.0
                if RxAngle > 360.0:
                    RxAngle = RxAngle - 360.0
                # Correct ULA with Rx orientation
                AoA_az = -RxAngle + AoA_az  # angle_new = - delta_axis + angle_wi;

                # first ray is the strongest, store its angles
                receiverPositions[s, r, 3] = AoD_el[0]
                receiverPositions[s, r, 4] = AoD_az[0]
                receiverPositions[s, r, 5] = AoA_el[0]
                receiverPositions[s, r, 6] = AoA_az[0]

                isLOSperRay = insiteData[:, 6]
                pathPhases = insiteData[:, 7]

                # in case any of the rays in LOS, then indicate that the output is 1
                isLOS = 0  # for the channel
                if np.sum(isLOSperRay) > 0:
                    isLOS = 1
                    numLOS += 1
                else:
                    numNLOS += 1
                receiverPositions[s, r, 7] = isLOS
                mimoChannel = getNarrowBandULAMIMOChannel(
                    AoD_az,
                    AoA_az,
                    gain_in_dB,
                    cfg["number_Tx_antennas"],
                    cfg["number_Rx_antennas"],
                    cfg["normalizedAntDistance"],
                    cfg["angleWithArrayNormal"],
                )
                # equivalentChannel = getDFTOperatedChannel(mimoChannel, number_Tx_antennas, number_Rx_antennas)
                # equivalentChannelMagnitude = np.abs(equivalentChannel)
                episodeOutputs[s, r] = np.abs(mimoChannel)
                allOutputs[count] = episodeOutputs[s, r]
                count += 1

            # finished processing this episode
        # Save beam per episode
        """
        npz_name = outputFolder + 'beams_output' + '_positions_e_' + str(e) + '.npz'
        np.savez(npz_name, episodeOutputs=episodeOutputs)
        np.savez(npz_name, receiverPositions=receiverPositions)
        print('Saved file ', npz_name) """

    # print("alloutputs shape = ", allOutputs.shape)
    mimoChannels_test = allOutputs[limit:]
    mimoChannels_train = allOutputs[:limit]
    print(
        "alloutputs shape = ",
        allOutputs.shape,
        "train = ",
        mimoChannels_train.shape,
        "test = ",
        mimoChannels_test.shape,
    )

    mimoChannels_train = np.array(mimoChannels_train).real
    mimoChannels_test = np.array(mimoChannels_test).real

    npz_name_train = cfg["outputFolder"] + "mimoChannels_train" + ".npz"
    np.savez(npz_name_train, output_classification=mimoChannels_train)
    print("Saved file ", npz_name_train)

    npz_name_validation = cfg[
        "outputFolder"] + "mimoChannels_validation" + ".npz"
    np.savez(npz_name_validation, output_classification=mimoChannels_test)
    print("Saved file ", npz_name_validation)

    print("Sanity check (must be 0 NaN) sum of isNaN = ",
          np.sum(np.isnan(allOutputs[:])))
    print("total numOfInvalidChannels = ", numOfInvalidChannels)
    print("total numOfValidChannels = ", numOfValidChannels)
    print("Sum = ", numOfValidChannels + numOfInvalidChannels)

    print("total numNLOS = ", numNLOS)
    print("total numLOS = ", numLOS)
    print("Sum = ", numLOS + numNLOS)
def processBeamsOutput(data_folder, dataset, limit):
    print("Generating Beams ...")
    # file generated with ak_generateInSitePlusSumoList.py:
    # need to use both LOS and NLOS here, cannot use restricted list because script does a loop over all scenes
    insiteCSVFile = dataset + "/" + "CoordVehiclesRxPerScene_s008.csv"
    numEpisodes = 2086  # total number of episodes
    outputFolder = data_folder + "/beam_output/"
    # parameters that are typically not changed
    inputPath = (
        dataset
        + "/"
        + "ray_tracing_data_s008_carrier60GHz/rosslyn_mobile_60GHz_ts0.1s_V_Lidar_e"
    )
    normalizedAntDistance = 0.5
    angleWithArrayNormal = 0  # use 0 when the angles are provided by InSite

    # ULA
    number_Tx_antennas = 32
    number_Rx_antennas = 8

    if not os.path.exists(outputFolder):
        os.makedirs(outputFolder)

    # initialize variables
    numOfValidChannels = 0
    numOfInvalidChannels = 0
    numLOS = 0
    numNLOS = 0
    count = 0

    """use dictionary taking the episode, scene and Rx number of file with rows e.g.:
    0,0,0,flow11.0,Car,753.83094753535,649.05232524135,1.59,D:/insitedata/noOverlappingTx5m/run00000,LOS=0
    0,0,2,flow2.0,Car,753.8198286576,507.38595866735,1.59,D:/insitedata/noOverlappingTx5m/run00000,LOS=1
    0,0,3,flow2.1,Car,749.7071175056,566.1905128583,1.59,D:/insitedata/noOverlappingTx5m/run00000,LOS=1"""

    with open(insiteCSVFile, "r") as f:
        insiteReader = csv.DictReader(f)
        insiteDictionary = {}
        numExamples = 0
        for row in insiteReader:
            isValid = row["Val"]  # V or I are the first element of the list thisLine
            if isValid == "V":  # filter the valid channels
                numExamples += 1
                thisKey = (
                    str(row["EpisodeID"])
                    + ","
                    + str(row["SceneID"])
                    + ","
                    + str(row["VehicleArrayID"])
                )
                insiteDictionary[thisKey] = row
        lastEpisode = int(row["EpisodeID"])
    allOutputs = np.nan * np.ones(
        (numExamples, number_Rx_antennas, number_Tx_antennas), np.complex128
    )

    for e in range(numEpisodes):
        # print("Episode # ", e)
        b = h5py.File(inputPath + str(e) + ".hdf5", "r")
        allEpisodeData = b.get("allEpisodeData")
        numScenes = allEpisodeData.shape[0]
        numReceivers = allEpisodeData.shape[1]
        # store the position (x,y,z), 4 angles of strongest (first) ray and LOS or not
        receiverPositions = np.nan * np.ones((numScenes, numReceivers, 8), np.float32)
        # store two integers converted to 1
        episodeOutputs = np.nan * np.ones(
            (numScenes, numReceivers, number_Rx_antennas, number_Tx_antennas),
            np.float32,
        )

        for s in range(numScenes):
            for r in range(numReceivers):  # 10
                insiteData = allEpisodeData[s, r, :, :]

                # if insiteData corresponds to an invalid channel, all its values will be NaN.
                # We check for that below
                numNaNsInThisChannel = sum(np.isnan(insiteData.flatten()))
                if numNaNsInThisChannel == np.prod(insiteData.shape):
                    numOfInvalidChannels += 1
                    continue  # next Tx / Rx pair
                thisKey = str(e) + "," + str(s) + "," + str(r)

                try:
                    thisInSiteLine = list(
                        insiteDictionary[thisKey].items()
                    )  # recover from dic
                except KeyError:
                    print("Could not find in dictionary the key: ", thisKey)
                    print("Verify file", insiteCSVFile)
                    exit(-1)
                # tokens = thisInSiteLine.split(',')
                if numNaNsInThisChannel > 0:
                    numOfValidRays = int(
                        thisInSiteLine[8][1]
                    )  # number of rays is in 9-th position in CSV list
                    # I could simply use
                    # insiteData = insiteData[0:numOfValidRays]
                    # given the NaN are in the last rows, but to be safe given that did not check, I will go for a slower solution
                    insiteDataTemp = np.zeros((numOfValidRays, insiteData.shape[1]))
                    numMaxRays = insiteData.shape[0]
                    validRayCounter = 0
                    for itemp in range(numMaxRays):
                        if (
                            sum(np.isnan(insiteData[itemp].flatten())) == 1
                        ):  # if insite version 3.2, else use 0
                            insiteDataTemp[validRayCounter] = insiteData[itemp]
                            validRayCounter += 1
                    insiteData = insiteDataTemp  # replace by smaller array without NaN
                receiverPositions[s, r, 0:3] = np.array(
                    [thisInSiteLine[5][1], thisInSiteLine[6][1], thisInSiteLine[7][1]]
                )

                numOfValidChannels += 1
                gain_in_dB = insiteData[:, 0]
                timeOfArrival = insiteData[:, 1]
                # InSite provides angles in degrees. Convert to radians
                # This conversion is being done within the channel function
                AoD_el = insiteData[:, 2]
                AoD_az = insiteData[:, 3]
                AoA_el = insiteData[:, 4]
                AoA_az = insiteData[:, 5]
                RxAngle = insiteData[:, 8][0]
                RxAngle = RxAngle + 90.0
                if RxAngle > 360.0:
                    RxAngle = RxAngle - 360.0
                # Correct ULA with Rx orientation
                AoA_az = -RxAngle + AoA_az  # angle_new = - delta_axis + angle_wi;

                # first ray is the strongest, store its angles
                receiverPositions[s, r, 3] = AoD_el[0]
                receiverPositions[s, r, 4] = AoD_az[0]
                receiverPositions[s, r, 5] = AoA_el[0]
                receiverPositions[s, r, 6] = AoA_az[0]

                isLOSperRay = insiteData[:, 6]
                pathPhases = insiteData[:, 7]

                # in case any of the rays in LOS, then indicate that the output is 1
                isLOS = 0  # for the channel
                if np.sum(isLOSperRay) > 0:
                    isLOS = 1
                    numLOS += 1
                else:
                    numNLOS += 1
                receiverPositions[s, r, 7] = isLOS
                mimoChannel = getNarrowBandULAMIMOChannel(
                    AoD_az,
                    AoA_az,
                    gain_in_dB,
                    number_Tx_antennas,
                    number_Rx_antennas,
                    normalizedAntDistance,
                    angleWithArrayNormal,
                )
                equivalentChannel = getDFTOperatedChannel(
                    mimoChannel, number_Tx_antennas, number_Rx_antennas
                )
                equivalentChannelMagnitude = np.abs(equivalentChannel)
                episodeOutputs[s, r] = np.abs(equivalentChannel)
                allOutputs[count] = episodeOutputs[s, r]
                count += 1

            # finished processing this episode
        # Save beam per episode
        """
        npz_name = outputFolder + 'beams_output' + '_positions_e_' + str(e) + '.npz'
        np.savez(npz_name, episodeOutputs=episodeOutputs)
        np.savez(npz_name, receiverPositions=receiverPositions)
        print('Saved file ', npz_name) """

    beams_output_test = allOutputs[limit:]
    beams_output_train = allOutputs[:limit]

    npz_name_train = outputFolder + "beams_output_train" + ".npz"
    np.savez(npz_name_train, output_classification=beams_output_train)
    print("Saved file ", npz_name_train)

    npz_name_validation = outputFolder + "beams_output_validation" + ".npz"
    np.savez(npz_name_validation, output_classification=beams_output_test)
    print("Saved file ", npz_name_validation)

    print(
        "Sanity check (must be 0 NaN) sum of isNaN = ", np.sum(np.isnan(allOutputs[:]))
    )
    print("total numOfInvalidChannels = ", numOfInvalidChannels)
    print("total numOfValidChannels = ", numOfValidChannels)
    print("Sum = ", numOfValidChannels + numOfInvalidChannels)

    print("total numNLOS = ", numNLOS)
    print("total numLOS = ", numLOS)
    print("Sum = ", numLOS + numNLOS)
Ejemplo n.º 5
0
                if numUserRays == 0:
                    Huser[iUser, :, :] = np.nan * np.zeros((1, Nr, Nt))
                else:
                    numValidChannels += 1
                    gain_in_dB = episodeRays[iScene, iUser, 0:numUserRays, 0]
                    timeOfArrival = episodeRays[iScene, iUser, 0:numUserRays,
                                                1]
                    AoD_el = episodeRays[iScene, iUser, 0:numUserRays, 2]
                    AoD_az = episodeRays[iScene, iUser, 0:numUserRays, 3]
                    AoA_el = episodeRays[iScene, iUser, 0:numUserRays, 4]
                    AoA_az = episodeRays[iScene, iUser, 0:numUserRays, 5]
                    isLOSperRay = episodeRays[iScene, iUser, 0:numUserRays, 6]
                    pathPhases = episodeRays[iScene, iUser, 0:numUserRays, 7]

                    Huser[iUser,:,:] = mimo_channels.getNarrowBandULAMIMOChannel(\
                        AoD_az, AoA_az, gain_in_dB, Nt, Nr)

            Ht[iEpisode, :, :, :, :] = Huser

    print('### Finished processing channels')
    print('\t %d Total of channels' % numChannels)
    print('\t %d Total of valid channels' % numValidChannels)

# permute dimensions before reshape: scenes before episodes
# found out np.moveaxis as alternative to permute in matlab
Ht = Ht[~np.isnan(Ht).any(axis=4)]
Ht = Ht.reshape((-1, 1, 1, Nr, Nt))

n_episodes, n_scenes, n_receivers, n_r, n_t = Ht.shape

Ht = np.moveaxis(Ht, 1, 0)