awgn

Add white Gaussian noise to signal

Syntax

Y = awgn(X,snr)

Y = awgn(X,snr,signalpower)

Y = awgn(X,snr,signalpower,randobject)

Y = awgn(X,snr,signalpower,seed)

Y = awgn(___,powertype)

[Y,var] = awgn(___)

Description

Y = awgn(X,snr) adds white Gaussian noise to the vector signal X. This syntax assumes that the power of X is 0 dBW. For more information about additive white Gaussian noise, see What is AWGN?

Y = awgn(X,snr,signalpower) accepts an input signal power value in dBW. To measure the power of X before adding noise, specify signalpower as 'measured'. The 'measured' option does not generate the requested average SNR for repeated awgn function calls in a loop if the input signal power varies over time due to fading and the coherence time of the channel is larger than the input duration.

Y = awgn(X,snr,signalpower,randobject) additionally accepts a random number stream object to generate normal random noise samples. For information about producing repeatable noise samples, see Tips.

Y = awgn(X,snr,signalpower,seed) specifies a seed value for initializing the normal random number generator that is used to add white Gaussian noise to the input signal.

Y = awgn(___,powertype) specifies the signal and noise power type as 'dB' or 'linear' in addition to the input arguments in any of the previous syntaxes. For information on the relationships between SNR and other measures of the relative power of the noise, such as E_s/N₀, and E_b/N₀, see AWGN Channel Noise Level.

[Y,var] = awgn(___) also returns the total noise variance used to produce random noise samples in linear scale.

Examples

Add AWGN to Sawtooth Signal

Create a sawtooth wave.

t = (0:0.1:60)';
x = sawtooth(t);

Add white Gaussian noise and plot the results.

y = awgn(x,10,'measured');
plot(t,[x y])
legend('Original Signal','Signal with AWGN')

Estimate Symbol Rate for General QAM Modulation in AWGN Channel

Transmit and receive data using a nonrectangular 16-ary constellation in the presence of Gaussian noise. Show the scatter plot of the noisy constellation and estimate the symbol error rate (SER) for two different SNRs.

Create a 16-QAM constellation based on the V.29 standard for telephone-line modems.

c = [-5 -5i 5 5i -3 -3-3i -3i 3-3i 3 3+3i 3i -3+3i -1 -1i 1 1i];
sigpower = pow2db(mean(abs(c).^2));
M = length(c);

Generate random symbols.

data = randi([0 M-1],2000,1);

Modulate the data by using the genqammod function. General QAM modulation is necessary because the custom constellation is not rectangular.

modData = genqammod(data,c);

Pass the signal through an AWGN channel with a 20 dB SNR.

rxSig = awgn(modData,20,sigpower);

Display a scatter plot of the received signal and the reference constellation c.

h = scatterplot(rxSig);
hold on
scatterplot(c,[],[],'r*',h)
grid
hold off

Demodulate the received signal by using the genqamdemod function. Determine the number of symbol errors and the SER.

demodData = genqamdemod(rxSig,c);
[numErrors,ser] = symerr(data,demodData)

numErrors = 
4

ser = 
0.0020

Repeat the transmission and demodulation process with an AWGN channel with a 10 dB SNR. Determine the SER for the reduced SNR. As expected, the performance degrades when the SNR is decreased.

rxSig = awgn(modData,10,sigpower);
demodData = genqamdemod(rxSig,c);
[numErrors,ser] = symerr(data,demodData)

numErrors = 
457

ser = 
0.2285

Repeatable AWGN Simulation

Generate random data symbols and the 4-PSK modulated signal.

M = 4;
k = log2(M);
snr = 3;
data = randi([0 M-1],2000,1);
x = pskmod(data,M);

Set the random number generator seed.

seed = 12345;

Generate repeatable random noise using the rng function before calling the awgn function.

rng(seed);
y = awgn(x,snr);

Compute the bit errors.

dataHat = pskdemod(y,M);
numErr1 = biterr(data,dataHat,k)

numErr1 = 
309

Reset the random number generator seed.

rng(seed);

Demodulate the PSK signal and compute the bit errors.

y = awgn(x,snr);
dataHat = pskdemod(y,M);
numErr2 = biterr(data,dataHat,k)

numErr2 = 
309

Compare numErr1 to numErr2. The errors are equal even after you reset the random number generator seed.

isequal(numErr1, numErr2)

ans = logical
   1

Repeatable AWGN with RandStream

Generate white Gaussian noise addition results by using a RandStream object and the reset object function.

Specify the input signal power of as 0 dBW, add noise to produce an SNR of 10 dB, and use a local random stream. Add white Gaussian noise to sigin two times to produce sigout1 and sigout2. Use isequal to compare sigout1 to sigout2. The outputs are not equal when you do not reset the random stream.

S = RandStream('mt19937ar',Seed=5489);
sigin = sqrt(2)*sin(0:pi/8:6*pi);
sigout1 = awgn(sigin,10,0,S);
sigout2 = awgn(sigin,10,0,S);
isequal(sigout1,sigout2)

ans = logical
   0

Reset the random stream object, returning the object to its state prior to adding AWGN to sigout1. Add AWGN to sigin to produce sigout3, and then compare sigout1 to sigout3. The outputs are equal when you reset the random stream.

reset(S);
sigout3 = awgn(sigin,10,0,S);
isequal(sigout1,sigout3)

ans = logical
   1

Input Arguments

X — Input signal

scalar | vector | numeric array | dlarray object

Input signal, specified as a scalar, vector, numeric array, or a dlarray (Deep Learning Toolbox) object. The power of the input signal is assumed to be 0 dBW. If X is complex, awgn adds a complex noise. For more information, see Array Support.

Data Types: double
Complex Number Support: Yes

snr — Signal-to-noise ratio

scalar | vector

Signal-to-noise ratio in dB, specified as:

• a scalar if the input signal is a scalar or a vector.

• a scalar or a vector if the input signal is a 3D array. For more information see, Array Support.

The function applies the same snr value to each channel. The columns of the input signal represent the different channels of a multichannel signal.

Data Types: double

signalpower — Signal power

scalar | 'measured'

Signal power in dBW, specified as a scalar or 'measured'.

• Scalar — The value is used as the signal level of X to determine the noise level required to achieve the specified snr.

• 'measured' — The signal level of X is computed to determine the noise level required to achieve the specified snr. You cannot use this value if the input signal is a dlarray or if snr is a vector.

If the input signal is a multichannel signal, the function calculates the signalpower value across all channels as a single value. It then uses the value to calculate the noise level for all the channels.

Data Types: double

randobject — Random number stream object

RandStream object

Random number stream object, specified as a RandStream object. The state of the random stream object determines the sequence of numbers produced by the randn function. Configure the random stream object using the reset (RandStream) function and its properties. You cannot specify a random stream object, if the input signal is a dlarray.

For information about producing repeatable noise samples, see Tips.

seed — Random number generator seed

scalar

Random number generator seed value, specified as a scalar. You cannot specify a seed, if the input signal is a dlarray.

Data Types: double

powertype — Signal power unit

'dB' (default) | 'linear'

Signal power unit, specified as 'dB' or 'linear'.

• When powertype is 'dB', snr is measured in dB and signalpower is measured in dBW.

• When powertype is 'linear', the snr is measured as a ratio and signalpower is measured in watts assuming a reference load of 1 ohms.

To set the powertype argument, you must also set snr and signalpower.

Output Arguments

Y — Output signal

scalar | vector | dlarray object

Output signal, returned as a scalar, vector, or a dlarray (Deep Learning Toolbox) object. The returned output signal is the input signal with added white Gaussian noise. If the input signal, X, is a dlarray, then this output, Y, is also a dlarray. For more information, see Array Support.

var — Total noise variance in linear scale

scalar | vector

Total noise variance in linear scale, returned as a positive scalar or a vector when snr is a vector. The function uses the noise variance to generate random noise samples.

More About

What is AWGN?

Additive white Gaussian noise (AWGN) is a simple noise model that represents electron motion in the RF front end of a receiver. As the name implies, the noise gets added to the signal. The noise is called white because it is spectrally flat across the entire sampling bandwidth. Analogously, the color white contains equal spectral power levels at all frequencies of the visible light spectrum. The noise is Gaussian because its amplitude can be modeled with a normal probability distribution.

The AWGN channel is often used to model a satellite communications channel, since that channel typically does not suffer from common terrestrial impairments like fading, multipath, and interference. An AWGN channel serves as a good starting point for the analysis of terrestrial wireless links because it establishes a best-case bound on the bit error rate performance of a terrestrial link.

Array Support

The awgn function supports input signals represented in a numeric array, dlarray (Deep Learning Toolbox), or gpuArray (Parallel Computing Toolbox).

The number of batch observations (N_B) is an optional dimension that can be added to these inputs for all supported data types.

• X — The input signal can be an array of up to three dimensions, specified as N_S-by-N_C-by-N_B array.

• snr can be a scalar or a vector of size N_B.

N_S is the number of samples and N_C is the number of channels.

For a list of Communications Toolbox™ features that support dlarray objects, see AI for Wireless.

Tips

• For information on the relationships between SNR and other measures of the relative power of the noise, such as E_s/N₀, and E_b/N₀, see AWGN Channel Noise Level.

• To generate repeatable white Gaussian noise samples, do one of the following:

  • Use rng(seed) before calling the awgn function to generate repeatable random noise.

  • Provide a static seed value as an input to awgn except when the input is a dlarray object.

  • Use the reset (RandStream) function on the randobject before passing it as an input to awgn except when the input is a dlarray object.

  • Provide randobject in a known state as an input to awgn except when the input is a dlarray object. For more information, see RandStream.

• To reproduce CPU random number streams on a GPU, the random number generator used on both must align. For more information, see Random Number Streams on a GPU (Parallel Computing Toolbox).

Extended Capabilities

C/C++ Code Generation

Generate C and C++ code using MATLAB® Coder™.

Usage notes and limitations:

• Code generation supported, except for syntaxes that include a RandStream object.

GPU Arrays

Accelerate code by running on a graphics processing unit (GPU) using Parallel Computing Toolbox™.

This function supports GPU array inputs. For more information, see Run MATLAB Functions on a GPU (Parallel Computing Toolbox).

Usage notes and limitations:

• When the input signal X is a gpuArray object, you cannot specify the seed and randobject input arguments and the snr input argument can not be set to "measured".

Version History

Introduced before R2006a

R2023b: Add deep learning array support

The awgn function adds support for dlarray (Deep Learning Toolbox) object processing for deep learning applications.

R2023b: Add GPU array support

The awgn function adds support for gpuArray (Parallel Computing Toolbox) object processing to run code on a graphics processing unit (GPU).

R2023b: awgn improves performance when adding complex noise to complex inputs

The awgn function improves performance when adding complex noise to complex inputs. For more information, see Match complexity with "like", and use "like" with RandStream object.

This function update results in numerical output changes due to different random numbers generated for the noise added to the signal. If you need to replicate pre-R2023b awgn output, review these code examples:

Before R2023b	Code to Reproduce Pre-R2023b Output
x = ones(2)+1i*ones(2); y_23a = awgn(x,5,1) y_23a = 2×2 complex 1.2399 + 1.1422i -0.0078 + 0.8066i 1.8182 + 0.4166i 1.3847 + 1.1529i	x = ones(2)+1i*ones(2); y_23b = complex(awgn(real(x),5,1 - 10*log10(2)), ... awgn(imag(x),5,1 - 10*log10(2))) y_23b = 2×2 complex 1.2399 + 1.1422i -0.0078 + 0.8066i 1.8182 + 0.4166i 1.3847 + 1.1529i
y_23a = awgn(x,1,'measured') y_23a = 2×2 complex 1.4792 + 1.2841i -1.0132 + 0.6136i 2.6345 - 0.1655i 1.7684 + 1.3054i	y_23b = complex(awgn(real(x),1,'measured'), ... awgn(imag(x),1,'measured')) y_23b = 2×2 complex 1.4792 + 1.2841i -1.0132 + 0.6136i 2.6345 - 0.1655i 1.7684 + 1.3054i
y_23a = awgn(x,5,1,'linear') y_23a = 2×2 complex 1.1700 + 1.1008i 0.2857 + 0.8629i 1.5799 + 0.5865i 1.2726 + 1.1083i	y_23b = complex(awgn(real(x),5,1/2,'linear'), ... awgn(imag(x),5,1/2,'linear')) y_23b = 2×2 complex 1.1700 + 1.1008i 0.2857 + 0.8629i 1.5799 + 0.5865i 1.2726 + 1.1083i

R2023b: awgn uses Mersenne Twister random number generator when seed value is specified as an input

When a seed value is specified in the awgn function call, the random number generator (RNG) now uses the Mersenne Twister RNG instead of the v5 normal RNG.

This function update results in numerical output changes due to different random numbers generated for the noise added to the signal. The updated awgn improves the added noise statistics, but if you need to replicate pre-R2023b awgn output, review these code examples:

Before R2023b	Code to Reproduce Pre-R2023b Output
x = ones(2); seed = 5; y_23a = awgn(x,1,1,seed) y_23a = 2×2 1.9876 0.6199 1.2590 1.1726	x = ones(2); stream = RandStream('shr3cong','Seed',5); y_23b = awgn(x,1,1,stream); y_23b = 2×2 1.9876 0.6199 1.2590 1.1726