Why do I Receive this error? Requested 708198x708198 (3736.8GB) array exceeds maximum array size preference (15.9GB). This might cause MATLAB to become unresponsive.

Hi, i'm trying to implement a function that simulates the effects of a Leslie Speaker. It actually does its job, but when I call it in the main environment (where I should also compute MSE), I get this error: Requested 708198x708198 (3736.8GB) array exceeds maximum array size preference (15.9GB). This might cause MATLAB to become unresponsive.

The problem is surely in my implementation of the Leslie function, as it's mandatory not to change the main script. Where is the issue? How can I solve it?

Here's the function:

function [y, y_lpf, y_hpf, y_hp_sdf] = leslie(x, Fs, freq)

% Length of the input signal

N = length(x);

% Global modulator parameters

alpha = 0.9;

% Tremble spectral delay filter parameter

Ms_t = 0.2;

Mb_t = -0.75;

N_sdf_t = 4;

% Bass spectral delay filter parameter

Ms_b = 0.04;

Mb_b = -0.92;

N_sdf_b = 3;

% Cross-over network design

fc = 800; % Cutoff frequency

% Compute the coefficients for the two 4th order Butterworth filters

[b_lp, a_lp] = butter(4, fc/(Fs/2), 'low'); % LPF design

[b_hp, a_hp] = butter(4, fc/(Fs/2), 'high'); % HPF design

% Allocate input and output buffers for IIR filters

% HP filter buffers

hpf.state = zeros(max(length(b_hp),length(a_hp))-1, 1);

hpf.in = zeros(max(length(b_hp),length(a_hp))-1, 1);

% LP filter buffers

lpf.state = zeros(max(length(b_lp),length(a_lp))-1, 1);

lpf.in = zeros(max(length(b_lp),length(a_lp))-1, 1);

% Treble SDF filter buffers

sdf_h.state = zeros(N_sdf_t, 1);

sdf_h.in = zeros(N_sdf_t, 1);

% Bass SDF filter buffers

sdf_b.state = zeros(N_sdf_b, 1);

sdf_b.in = zeros(N_sdf_b, 1);

% Modulators

m_b = Ms_b * sin(2*pi*freq*(0:N-1)/Fs) + Mb_b; % Bass modulator

m_t = Ms_t * sin(2*pi*(freq+0.1)*(0:N-1)/Fs) + Mb_t; % Tremble modulator

% Sample processing

y = zeros(N, 1);

y_lpf = zeros(N, 1);

y_hpf = zeros(N, 1);

y_hp_sdf = zeros(N, 1);

for n = 1:N

% Compute crossover network filters outputs

y_lpf(n) = filter(b_lp, a_lp, x(n)) - lpf.in(1);

y_hpf(n) = filter(b_hp, a_hp, x(n)) - hpf.in(1);

% Compute bass SDF output

y_lp_sdf(n) = (1 + m_b(n)) * (y_lpf(n) - sdf_b.in(N_sdf_b));

% Compute treble SDF output

y_hp_sdf(n) = (1 + m_t(n)) * (y_hpf(n) - sdf_h.in(N_sdf_t));

% Implement AM modulation blocks

y_lp_am = (1 + alpha * m_b(n)) * y_lpf(n);

y_hp_am = (1 + alpha * m_t(n)) * y_hpf(n);

y(n) = y_lp_am + y_hp_am; % Output sample

end

end

Here's the main script:

%------------------------------------------%

% *** SSSP - HOMEWORK #3 *** %

%------------------------------------------%

% Emulation of the Leslie Speaker %

%------------------------------------------%

% Name: Gian Marco Ricci / Adriane Replogle %

% Student ID: 10660246 / 10646402 %

%------------------------------------------%

clear; close all; clc;

%% modulation speed

mod_speed = 'tremolo';

%% Read the input file

[x, Fs] = audioread('HammondRef.wav');

x=x(:,1); % take left channel only

%% FX parameters

switch lower(mod_speed)

case {'chorale'}

freq=2;

case {'tremolo'}

freq=6;

otherwise

error('mod_speed \"%s\" not found.', mod_speed)

end

%% Apply FX

y = leslie(x, Fs, freq);

%% Avoid any (possible) clipping

y = rescale(y,-1.,1.);

%% Playback

%audiowrite([mod_speed,'.wav'], y, Fs);

soundsc(y, Fs)

%% Read the reference audio file

dir_name = 'Leslie_ref';

addpath(dir_name);

[y_ref, ~] = audioread(fullfile(dir_name, strcat(mod_speed,'.wav')));

%% Display the MSE

MSE = mean(abs(y.'-y_ref).^2);

MSE_str = sprintf('MSE: %g', MSE);

disp(MSE_str)

Answer 1

[chicken vector]

The error message is self-explanatory. Your implementation might be correct but at some point you are generating a variable of size 708198x708198 that occupies mroe than 3 TB of space, which is ay over Matlab's limit (15.9GB), as specified by the error.

You need to restructure our data to decrease the size of you variable.

If you share more details about your code such as the input files and at what line the error happens it would be much easier to help you.

[Stephen23]

"which is ay over Matlab's limit (15.9GB)"

The limit actually depends on the largest contiguous memory provided by the OS. If available, MATLAB can use 100% of the installed memory. See "MaxPossibleArrayBytes" here:

https://www.mathworks.com/help/matlab/ref/memory.html

The simple point: it is not MATLAB's limit, but the user's own installed memory/OS limit.

[chicken vector]

I was trying to keep things simple, but thank you for precising that.

[Gian Marco Ricci]

The error is in mainHW3 (line 49)

MSE = mean(abs(y.'-y_ref).^2);

you'll also find the input file HammondRef.wav in the attached .zip below. It's used to compute x in mainHW3 line 16. X's dimensions are 708198x1 (same as y)

Let me know if you can help me!

[Torsten]

Most probably, y.' is a row and y_ref is a column vector or vice versa such that y.'-y_ref becomes a matrix of size 708198 x 708198. Either make both row or column vectors to continue.

[chicken vector]

When you do:

mean(abs(y.'-y_ref).^2);

you are transposing the first vector so the difference:

y.'-y_ref

is a matrix, rather than a vector, as you can tell from the example below:

ones(3,1) - ones(1,3)

ans = 3×3

0 0 0 0 0 0 0 0 0

If this was intended you have no other option that to change your code because you have a matrix of 49x10^10 elements and these are too much to handle.

Otherwise do:

y - y_ref