function out = RX_PN(in,dBc, cutoff, floor, Fs)
% =========================================================================
% Book title: Digital Front-End Compensation for Emerging Wireless Systems
% Authors: Fran鏾is Horlin (ULB), Andr� Bourdoux (IMEC)
% Editor: John Wiley & Sons, Ltd
% Date: 2008
% =========================================================================
%
% DESCRIPTION
% This function applies phase noise to the signal stream(s). Each row
% contains one stream. The same phase noise is applied to each stream as
% if they are working with the same local oscillator. The phase noise has
% a PSD defined by four values: integrated phase noise, cutoff frequency, a
% phase noise floor and the sample rate. At the cutoff frequency, the PSD
% starts a -20dB/dec roll-off
%
% INPUTS
% - in [length x Ns]: input signal where each row is a signal
% from a different antenna.
% - dBc [1]: integrated phase noise
% - cutoff [1]: cutoff frequency of the PSD
% - floor [1]: phase noise floor
% - Fs [1]: sample rate
%
% OUTPUTS
% - out [length x Ns]: Ns streams of output signals
[M N] = size(in);
ltx = 2^(fix(log2(M-0.5))+2); % ltx will always be at least 2xM
if cutoff/(Fs/ltx)<16,
ltx = 16*Fs/cutoff;
ltx = 2^(fix(log2(ltx-0.5))+1);
end
PhaseNoise = FreqSynthLorenzian_new(dBc, cutoff, floor, Fs, ltx).';
Nphn = length(PhaseNoise);
if Nphn<M
error(['Phase noise vector must be longer than ' num2str(Nsamples)])
end
Nstart = fix(rand(1,1)*(Nphn-M)); % random staring point
PhaseNoise = PhaseNoise(Nstart:Nstart+M-1);
% Add phase noise to data
out = zeros(size(in));
for k=1:N
out(:,k) = in(:,k).*PhaseNoise.';
end
%%%%%%%%%%%%%
function y = FreqSynthLorenzian_new( K, B, p2, Fs, ltx);
% - lo: time domain vector, possible time domain representation of the
% oscillator amplitude.
% - K: total noise power in dBc
% - B: 3dB bandwidth
% - p2: noise plateau level in dBc/Hz, to be set at #20dB above the noise
% floor.
% - ltx: length of the signal lo vector to return
% The negative frequencies of the BB equivalent LO spectrum are the
% complex conjugate of the positive frequencies (as a result, phi(t) is
% real). The reason for this is not the need for the phi(t) or lo(t) to be
% real, but it is coming from the analogy to FM or PM modulations that do
% create such 抯ymetric� frequency responses.
% - PHI(f), hence LO(f), should not be generated as a spectrum with a wanted
% shape, but as a PSD (representation for a random process) with a wanted
% mask.
global d
Ns = ltx; % Number of samples for ifft calculation
df = Fs/Ns; % frequency resolution
k = [0:1:Ns/2-1, -Ns/2:1:-1]; % frequency index range
% frequency range: f = k * df
p2o = p2;
p2 = 10^(p2/10); % in V^2/Hz
p2 = p2*df; % in V^2/df
Ko = K;
K = 10^(K/10); % in V^2
B = B/df; % 3-dB bandwidth index
% Low frequency part is defined by Lorenzian function
SSBmask = sqrt( K*B/pi./([1:Ns/2].^2+B^2) );
% High frequency part is defined by the noise floor of the system p2
SSBmask = max(SSBmask, sqrt(p2)*ones(size(SSBmask)));
%-- Phase noise PHI(f, f>0) is first generated as a wide band signal
PHI = sqrt(0.5) * abs( randn(1,Ns/2) + j*randn(1,Ns/2) );
PHI = PHI .* exp(j*2*pi*rand(1,Ns/2));
%-- Phase noise PHI(f, f>0) is shaped according to the wanted mask
PHI = PHI .* SSBmask;
%-- Phase noise PHI(f) is then generated from PHI(f, f>0)
% (no phase noise on the carrier)
PHI = [0 PHI(1:Ns/2-1) conj(PHI(Ns/2:-1:1))];
NoisePower = 10*log10( sum(abs(PHI).^2) );
if (0)
f = k(2:Ns/2)*df;
PHI_f = 20*log10(abs(PHI(2:Ns/2))/sqrt(df));
SSBmask_f = 20*log10(SSBmask(1:Ns/2-1)/sqrt(df));
% normalisation to get PHI(f) in dBc/Hz, and not dBc/df
figure(20),semilogx(f, PHI_f, 'b-','linewidth',1);
hold on; grid on; zoom on;
semilogx(f, SSBmask_f, 'r-','linewidth',2);
axis([f(1) f(end) p2o-10 SSBmask_f(1)+10]);
end;
%-- Correction for the integrated phase noise power:
PHI = PHI * 10^((Ko-NoisePower)/20);
%-- Phase noise phi(t) in the time domain
phi = ifft(PHI,Ns)*Ns;
%-- Local oscillator signal lo(t) in the time domain
lo = exp(j*phi);
%-- Local oscillator signal LO(f) in the frequency domain
if (0)
LO = fft(lo,Ns)/Ns;
f = k(2:Ns/2)*df;
LO_f = 20*log10(abs(LO(2:Ns/2))/sqrt(df));
SSBmask_f = 20*log10(SSBmask(1:Ns/2-1)/sqrt(df));
figure(21);semilogx(f, LO_f, 'k-','linewidth',1);
hold on; grid on; zoom on;
semilogx(f, SSBmask_f, 'r-','linewidth',2);
axis([f(1) f(end) p2o-10 SSBmask_f(1)+10]);
end;
%-- Prepare lo(t) for efficient use in dbbm_fe
y = lo(1,1:ltx*fix(Ns/ltx));
y = reshape(y,ltx,fix(Ns/ltx));