I would just multiply the entire signal by some constant ‘K’ less than 1:
out = median(signal) + (signal-median(signal))*K;
Example:
t = linspace(0, 2*pi);
signal = pi + sin(t) + 0.2*randn(size(t));
K = 0.7;
out = median(signal) + (signal-median(signal))*K;
figure(1)
plot(t, signal, t, out)
grid
