TOOL_overview.md

NOMA C++ Overview

NOMA on GPUs library toolset overview is given here.

NOMA offline data format

We have defined the following offline data content. This is used for scripting as well as for unit testing and debugging.

reference data or random seed value

This data is used in any tool you find the -r or --reference-data option e.g. in NOMA_demo and NOMA_cli tool as input or in NOMA_dataset as output.

Value	Format	Description	Comment
2	int32	Num keys	total keys
8	int32	Len name 1	key 1 name length
[...]	char[]	[mod_data]	key 1 name
12	int32	Len name 2	key 2 name length
[...]	char[]	[mod_training]	key 2 name
3	int32	Dimensions	num dims for key 1
3840	int32	Dim 1	dim 1 length
16	int32	Dim 2	dim 2 length
2	int32	Dim 3	dim 3 length
[...]	float[]	Data: 122880	3840 x 16 x 2
3	int32	Dimensions	num dims for key 2
685	int32	Dim 1	dim 1 length
16	int32	Dim 2	dim 2 length
2	int32	Dim 3	dim 3 length
[...]	float[]	Data: 21920	685 x 16 x 2

Example:

• cpp/data/offline/tx/NOMA_signals_qam16_complex.bin

mod_data: [3840, 8, 2]
mod_training: [685, 8, 2]

synchronized data

This data is used in any tool you find the -s or --synchronized-data option e.g. in NOMA_tool as input or in NOMA_convert as output.

Value	Format	Description	Comment
2	int32	Num keys	total keys
13	int32	Len name 1	key 1 name length
[...]	char[]	[rxSigTraining]	key 1 name
9	int32	Len name 2	key 2 name length
[...]	char[]	[rxSigData]	key 2 name
3	int32	Dimensions	num dims for key 1
685	int32	Dim 1	dim 1 length
16	int32	Dim 2	dim 2 length
2	int32	Dim 3	dim 3 length
[...]	float[]	Data: 21920	685 x 16 x 2
3	int32	Dimensions	num dims for key 2
3840	int32	Dim 1	dim 1 length
16	int32	Dim 2	dim 2 length
2	int32	Dim 3	dim 3 length
[...]	float[]	Data: 122880	3840 x 16 x 2

• cpp/data/offline/rx/time/rxData_QAM16_alltx_converted.bin
• cpp/data/offline/rx/ofdm/rxData_QAM16_alltx_converted.bin

rxSigTraining: [685, 16, 2]
rxSigData: [3840, 16, 2]

NOMA_detect

This tool runs the training and detection algorithms. The algorithm configuration is passed in JSON format either as string or in a file.

Optional arguments:
-h  --help                       shows help message and exits [default: false]
-v  --version                    prints version information and exits [default: false]
-s  --synchronized-data          synchronized data file containing received training and data signals [required]
-r  --reference-data             reference data file
-rs --random-seed                random seed for generation of data (this overrides the --reference-data argument)
-d  --detected-data              detected (estimated) data file
-cf --config-file                filename for JSON file containing the algorithm configuration
-cs --config-string              JSON string containing the algorithm configuration [default: "{ "algorithm": { "otype": "APSM" } }"]
-o  --output-mode                print mode for error metric (BER = bit errors, SER = symbol errors, EVM = Error Vector Magnitude, scripting = disable all output except "ber,ser,evm") [default: "SER"]
-th --train-history-file         filename for training history CSV-file (test set will be used as validation set) [default: ""]
-p  --parallel-processing        process all users in parallel [default: false]

JSON config format

{
    "algorithm": { ... },  // configuration for detection algorithm (see below)
    "modulation": "QPSK",  // modulation used for data generation and error computation (valid values: off, bpsk, qpsk, qam16, qam64, qam256, qam1024, qam4096, qam16384)
    "user": 0,             // user, which should be decoded
    "antennas": { ... }    // selection of antennas (see below)
}

Algorithms

LLS

Linear least squares algorithm

{
    "otype": "LLS"
}

APSM

Adaptive projected subgradient method

{
    "otype": "APSM",
    "llsInitialization": false,      // should the linear weights be initialized with the LLS solution?
    "trainingShuffle": false,        // should the training data be shuffled?
    "gaussianKernelWeight": 0.5,     // Gaussian Kernel Weight
    "gaussianKernelVariance": 0.05,  // Gaussian Kernel Variance
    "windowSize": 20,                // number of past samples to reuse for each training iteration
    "sampleStep": 2,                 // number of samples the training window will advance in each iteration
    "trainPasses": 1,                // number of training passes over the training data
    "startWithFullWindow": false,    // if true skip the smaller windows at the beginning of training
    "dictionarySize": 685,           // maximum number of (complex) basis vectors to be saved in the dictionary
    "normConstraint": 0.,            // norm constraint to use for dictionary sparsification (set to 0 to disable)
    "eB": 0.001,                     // hyperslab width
    "detectVersion": "balanced"      // which (optimized) implementation to use (valid values: original, oldfast, shmem, balanced)

}

Antennas

All Antennas

{
    "otype": "all"
}

Direct Pattern

{
    "otype": "pattern",
    "pattern": "0011100110111010"  // directly specify which antennas should be used (should contain 16 chars which can be 1 (on) or 0(off))
}

Equidistant

{
    "otype": "equidistant",
    "number": 16 // number of antennas which will be chosen equidistantly
}

Random

{
    "otype": "random",
    "number": 16 // number of antennas which will be chosen randomly
}

First

{
    "otype": "first",
    "number": 16 // number of antennas which will be chosen from the beginning
}

Examples

Run the NOMA algorithm:

cd build

bin/NOMA_detect -r dataset/offline/tx/NOMA_signals_qpsk_complex.bin -s dataset/offline/rx/time/rxData_QPSK_alltx_converted.bin -cs '{"algorithm": {"otype":"APSM", "windowSize": 80}, "modulation":"QPSK", "antennas": {"otype": "equidistant", "number": 8}, "user": 1}'
bin/NOMA_detect -r dataset/offline/tx/NOMA_signals_qpsk_complex.bin -s dataset/offline/rx/ofdm/rxData_QPSK_alltx_converted.bin -cs '{"algorithm": {"otype":"APSM", "windowSize": 80}, "modulation":"QPSK", "antennas": {"otype": "equidistant", "number": 8}, "user": 1}'