5G Signal Pro-

cessing (EE45GS)
Report on OFDM vs GFDM

Dan Serbanoiu S.N. 200241287

 Abstract
The given executable implements a simulation of OFDM and GFDM transmission/reception and
was coded and tested using the MATLAB R2020b environment. The program allows the user to
change the outcome of the simulation by editing the parameters of “ofdm” and “gfdm” using the
structures (.p1) and (.p2), preemptively initialized with a default set of values. The simulation en-
codes random symbols to OFDM and GFDM signals, sends them through a AWGN channel af-
fected by different SNR values and decodes the received signals, producing the corresponding
figures of spectrum, eye-diagrams, constellation diagrams, BER, SER, Q-Factor. The script then
saves the figures in a local folder called “figures”.
Algorithm
close all;
clear;
if not(isfolder("figures"))
mkdir("figures")
end
%% (.p1) parameter sets for GFDM
gfdm = struct;
gfdm.K = 512; % Number of samples per sub-symbol
gfdm.M = 15; % Number of sub-symbols
gfdm.gfdmM = gfdm.M; % Accounts for OFDM simulation
gfdm.blockLength = gfdm.M*gfdm.K; % Block Length
gfdm.CP = 0.1; % Percentage of cyclic prefix
gfdm.pulse = 'rc'; % Pulse shaping filter (“rc” is raised cosine)
gfdm.a = 0.1; % Roll off factor of the pulse shaping filter
gfdm.mu = 4; % Modulation order of the QAM symbol
gfdm.subcarriers = 1:201; % Allocated subcarriers
gfdm.subsymbols = 2:15; % Allocated sub-symbols
gfdm.blocks = 10; % Number of blocks
%% (.p2) parameter sets for OFDM
ofdm = struct;
ofdm.K = 512;
ofdm.M = 1;
ofdm.gfdmM = gfdm.M;
ofdm.blockLength = ofdm.K;
ofdm.CP = 0.1;
ofdm.pulse = 'rc_td'; % (“rc_td” is a rectangular filter in frequency domain)
ofdm.a = 0.1;
ofdm.mu = 4;
ofdm.subcarriers = 1:201;
ofdm.subsymbols = 1;
ofdm.blocks = gfdm.blocks;
%% (.s1) Simulate GFDM and OFDM transmission
tic
[GFDMS, sym_gfdm_tx, cpx_gfdm_sym_tx, g1] = tx_simul(gfdm);
execution_tx_GFDM = toc;
tic
[OFDMS, sym_ofdm_tx, cpx_ofdm_sym_tx, g2] = tx_simul(ofdm);
execution_tx_OFDM = toc;
%% (.f1) Plot the PSD of the GFDM/OFDM signals
f = figure("Name", "PSD");
length = min(length(OFDMS), length(GFDMS));
t = linspace(-gfdm.K/2, gfdm.K/2, 2*length+1);
t = t(1:end-1)';
plot(t, mag2db(fftshift(abs(fft(OFDMS(1:length), 2*length))))/2, 'b');
hold on;
plot(t, mag2db(fftshift(abs(fft(GFDMS(1:length), 2*length))))/2, 'r');
ylim([-40, 30]);
xlabel('f/F'); ylabel('PSD [dB]');
legend({'OFDM', 'GFDM'});
print(['figures' filesep 'PSD_gfdm_ofdm'], "-dpng");
set(f, 'visible', 'off');
%% (.s2) Simulate reception with different SNR values
snr = 1:21;
gfdm_ber = []; gfdm_ser = []; ofdm_ber = []; ofdm_ser = [];
execution_rx_GFDM = []; execution_rx_OFDM = [];
[theory_ber, theory_ser] = berawgn(snr, "qam", gfdm.mu^2);
for sii=snr
tic
[sym_rx1, cpx_sym_rx1, ber1, ser1] = rx_simul(gfdm, awgn(GFDMS, sii), sym_gfdm_tx, g1);
execution_rx_GFDM = [execution_rx_GFDM; toc];
tic
[sym_rx2, cpx_sym_rx2, ber2, ser2] = rx_simul(ofdm, awgn(OFDMS, sii), sym_ofdm_tx, g2);
execution_rx_OFDM = [execution_rx_OFDM; toc];
gfdm_ber = [gfdm_ber; ber1]; gfdm_ser = [gfdm_ser; ser1];
ofdm_ber = [ofdm_ber; ber2]; ofdm_ser = [ofdm_ser; ser2];
%% (.f2) Plot the Eye Diagrams/Scatter Plots for different SNR value
if mod(sii, 3) == 0
print_eye_scatter("gfdm", gfdm, cpx_sym_rx1, sii);
print_eye_scatter("ofdm", ofdm, cpx_sym_rx2, sii);
end
end
%% (.f3) Plot the BER/SER values of the GFDM/OFDM signals
print_ber_ser_q('SER', snr, theory_ser, gfdm_ser, ofdm_ser);
print_ber_ser_q('BER', snr, theory_ber, gfdm_ber, ofdm_ber);
print_ber_ser_q('QFactor', snr, [], 0.5*erfcinv(2*gfdm_ber), 0.5*erfcinv(2*ofdm_ber));
% (.es) Function that can print Eye Diagram, Scatter plot of different formats
function print_eye_scatter(name, opt, sym, sii)
f1 = eyediagram(sym(1:800), ceil(sqrt(opt.mu)));
print(strcat('figures', filesep, name, '_eye_snr_', string(sii)), "-dpng");
f2 = scatterplot(sym, 1, 0, 'r.');
hold on;
 scatterplot(qammod(0:opt.mu^2-1, opt.mu^2, 'gray'), 1, 0, 'k*', f2);
title(strcat('Scatter plot,', {' '}, name, {', '}, 'SNR=', string(sii), 'dB'));
print(strcat('figures', filesep, name, '_scatter_snr_', string(sii)), "-dpng");
set(f1, 'visible', 'off');
set(f2, 'visible', 'off');
end
% (.bsq) Function that can print BER, SER, Q-VALUE of different formats
function print_ber_ser_q(name, snr, theory, gfdm, ofdm)
f = figure('Name', name);
if (size(theory) > 0)
semilogy(snr, theory,'-x');
hold on;
end
semilogy(snr, gfdm, 'o');
hold on;
semilogy(snr, ofdm, '*');
if (size(theory) > 0)
legend('theory', 'gfdm', 'ofdm');
else
legend('gfdm', 'ofdm');
end
xlabel('SNR, dB');
ylabel(name);
xlim([1, 20]);
print(['figures' filesep name], "-dpng");
set(f, 'visible', 'off');
end
% (.t) Simulation of transmitter GFDM/OFDM
function [signal, sym_tx, cpx_sym_tx, g] = tx_simul(opt)
signal = []; sym_tx = []; cpx_sym_tx = [];
% (.t1) Generate the transmitter pulse g
if strcmp(opt.pulse, 'rc')
t = linspace(-opt.M/2, opt.M/2, opt.M*opt.K + 1);
t = t(1:end-1);
t = t';
g = (sinc(t) .* cos(pi*opt.a*t) ./ (1-4*opt.a*opt.a*t.*t));
g = fftshift(g);
g(opt.K+1:opt.K:end) = 0;
g = g / sqrt(sum(g.*g));
elseif strcmp(opt.pulse, 'rc_td')
g = zeros(opt.M*opt.K, 1);
g(1:opt.K) = 1;
g = g / sqrt(sum(g.*g));
end
for b = 1:opt.blocks
gfdmM = 1;
 % (.t2) OFDM case
if opt.M == 1
gfdmM = opt.gfdmM;
end
for m = 1 : gfdmM
% (.t3) Generate random data and map to QAM constellation
sym = randi(2 ^ opt.mu, length(opt.subsymbols) * length(opt.subcarriers), 1) - 1;
cpx_sym = qammod(sym, 2^opt.mu, 'gray')/ sqrt(2/3 * (2^opt.mu - 1));
sym_tx = [sym_tx; sym];
cpx_sym_tx = [cpx_sym_tx; cpx_sym];
% (.t4) Map data to correct dimensions or pad it with zeros
if length(opt.subcarriers)== opt.K && length(opt.subsymbols) == opt.M
D = reshape(cpx_sym, opt.K, opt.M);
else
Dm = reshape(cpx_sym, length(opt.subcarriers), length(opt.subsymbols));
res1 = zeros(opt.K, length(opt.subsymbols));
res1(opt.subcarriers, :) = Dm;
res = zeros(opt.K, opt.M);
res(:, opt.subsymbols) = res1;
D = res;
end
% (.t5) Modulate in time
DD = repmat(opt.K*ifft(D), opt.M, 1);
x = zeros(opt.K*opt.M, 1);
for m=1:opt.M
symbol = DD(:, m) .* g;
symbol = circshift(symbol, opt.K*(m-1));
x = x + symbol;
end
% (.t6) Add CP
cp = ceil(opt.CP*opt.blockLength);
xcp = [x(end-cp + (1:cp), :); x];
signal = [signal; xcp];
end
end
end
% (.r) Simulation of receiver GFDM/OFDM
function [sym_rx, cpx_sym_rx, ber, ser] = rx_simul(opt, signal, sym_tx, g)
sym_rx = []; cpx_sym_rx = []; ber = 0; ser = 0;
for b = 1:opt.blocks
cp = ceil(opt.CP*opt.blockLength);
gfdmM = 1;
dim1 = opt.blockLength + cp;
dim2 = dim1;
% (.r1) OFDM case
if opt.M == 1
 gfdmM = opt.gfdmM;
dim1 = opt.gfdmM*opt.K + opt.gfdmM*cp;
dim2 = opt.K + cp;
end
for m = 1 : gfdmM
block = signal((b-1)*dim1 + (m-1)*dim2 + (1:dim2));
% (.r2) Remove CP
block = block(cp + 1:end);
% (.r3) Calculate the transmitter pulse g (Matched Filter)
g = g(round(1:opt.K/opt.K:end));
G = conj(fft(g));
% (.r4) Demodulate in frequency
L = length(G) / opt.M;
Xhat = fft(block);
Dhat = zeros(opt.K, opt.M);
for k=1:opt.K
carrier = circshift(Xhat, ceil(L*opt.M/2) - opt.M*(k-1));
carrier = fftshift(carrier(1:L*opt.M));
carrierMatched = carrier .* G;
dhat = ifft(sum(reshape(carrierMatched, opt.M, L), 2)/L);
Dhat(k,:) = dhat;
end
% (.r5) Unmap
if length(opt.subcarriers) == opt.K && length(subsymbols) == opt.M
dhat_mf = reshape(Dhat, numel(Dhat), 1);
else
Dm = Dhat(opt.subcarriers, opt.subsymbols);
dhat_mf = reshape(Dm, numel(Dm), 1);
end
% (.r6) Map constellation points to data
dhat_mf = dhat_mf * sqrt(2/3 * (2^opt.mu - 1));
shm = qamdemod(dhat_mf, 2^opt.mu, 'gray');
cpx_sym_rx = [cpx_sym_rx; dhat_mf];
sym_rx = [sym_rx; shm];
end
end
% (.r7) Calculate BER and SER
[dim1, dim2] = size(sym_tx);
sym_b_tx = de2bi(sym_tx, opt.mu); sym_b_rx = de2bi(sym_rx, opt.mu); err = [];
for ii = 1:size(sym_b_tx)
err = [err; nnz(xor(sym_b_tx(ii, :), sym_b_rx(ii, :)))];
end
ser = double(nnz(sym_tx - sym_rx)/dim1);
ber = sum(err)/(dim1*4);
end
 Editable Parameters
Parameter Description Realistic Values
K Samples per sub-symbol 64, 128, 512, 1024
M Sub-symbols 4, 5, 15 for GFDM, 1 for OFDM
CP Cyclic Prefix 0.1, 0.2, 0.3, 0.4, 0.5
pulse Pulse shaping filter rc = Raised cosine;
rc_td = Rectangular filter in fre-
quency domain
a Roll-off factor of the pulse shaping filter Between 0 and 1
mu Modulation of QAM symbol 2, 4, 8, 16
subcarriers Allocated sub-carriers 1:K
subsymbols Allocated sub-symbols 1:M for GFDM, 1 for OFDM
blocks Number of blocks 1, 2, 3, …, ∞
Results of the simulation
The simulation was executed with the following parameters:
GFDM: K = 512, M = 15, CP = 0.1, pulse = ‘rc’ (Raised Cosine filter), a = 0.1, mu = 4, blocks = 10,
subcarriers = 1:201, subsymbols = 2:15;
OFDM: K = 512, M = 1, CP = 0.1, pulse = ‘rc_td’ (Rectangular filter in frequency domain), a = 0.1,
mu = 4, subcarriers = 1:201, subsymbols = 1, blocks = 10.

Without CP With 10% CP

 OFDM/GFDM Components
Transmitter

o (.t1): The pulse shaping filter is generated according to the parameters specified in (.p1)
and (.p2);
o (.t2): The OFDM encoding is realized by the same function (.t) used for the GFDM encod-
ing. This is because OFDM is a special case of GFDM where M=1, every subcarrier is used
and the pulse shaping filter is rectangular in frequency domain. The following figure
shows how the OFDM encoding normally works:

o (.t3): A random set of symbols is generated for every block and mapped to QAM symbols.
The number of symbols is the number of chosen sub-carriers times the number of chosen
sub-symbols;
o (.t4): Because not all the sub-carriers and sub-symbols are used it is necessary to pad the
generated matrix with zeros in order to ensure that the encoding process works properly
o (.t5): A set of sub-carriers K and time slots M are encoded into the x[n] pulse through the
following pulse shaping operation:

o (.t6): The cyclic prefix is added to the pulse in time domain.

 Matched Filter Receiver

o (.r1): This is a list of operations that account for the OFDM decoding, that is done using (.r)
for the same reason explained in (.t2). The OFDM decoding normally is done as follows:

o (.r2): The cyclic prefix is removed from the received pulse;

o (.r3): The filter necessary for decoding is calculated from the filter used at the transmitter;
o (.r4): The matched filter operation that recovers the symbols in frequency domain is:

o (.r5): The padding is removed from the resulting matrix and it is converted into serial form;
o (.r6): The guessed constellation points are mapped to symbols according to the appropri-
ate decision boundaries.
OFDM GFDM
Advantages Advantages
o Subcarriers are orthogonal and their spectral o Each GFDM block is made of K subcarriers
energy does not interfere with the system’s and M time slots. To each carrier and each
ability to recover the original signal. Orthog- time slots corresponds one symbol. The
onality is achieved efficiently using the in- spread of the resulting signal in time do-
verse Fourier transform (IFFT), that maps N main results into a compression in fre-
symbols to a sum of N sinusoids in time do- quency domain achieving better spectral
main that correspond to N orthogonal sub- efficiency;
carriers in frequency domain; o Tail biting is employed to remove the
o The receiver can decode the received signal need for additional guard intervals and to
back to frequency domain efficiently using a provide an efficient FFT implementation;
Fourier transform (FFT); o The insertion of a cyclic prefix in larger
o The Fourier transformations ensure the peri- blocks increases further spectral efficiency,
odicity of the waveform which reduces inter- while still allowing for efficient channel
ference. For example, a pulse shifted in time, equalization in the frequency domain;
corresponds in frequency domain to a fre- o There is flexibility as it is possible to set the
quency proportional phase shift/rotation. block dimensions appropriately and the
pulse shaping filter based on the transmis-
sion scenario.
 Weaknesses Weaknesses
o The energy of the sidelobes of the compo- o It has a worse BER compared to OFDM if a
nents in frequency domain create interfer- matched filter is used at the receiver;
ence for adjacent subcarriers; o The transmitter and receiver are much
o The guard bands or the cyclic prefix can more complex to implement than OFDM
waste a lot of bandwidth; and the encoding process is slower.
o All the blocks employ the same subcarrier
spacing. There is no flexibility for different
applications;
o Synchronization is problematic.
5G standards, aims and implementation challenges
5G aims to implement network architectures that are programmable, flexible, have lower com-
plexity and more efficient operations and allow for faster development and deployment of ser-
vices. These networks need to be able to support richer services in many different sectors with
different requirements such as gaming, health care, autonomous ride-hailing etc. 5G achieves
programmability through VNF (virtualization on network functions), SDN (Software defined net-
working) and network slicing, allowing operators to manage and orchestrate their networks at
different levels (core network, access network and UE). In general, 5G aims to support high data
rates, low latency (1ms), mobility, connection density, energy efficiency, spectrum efficiency and
area traffic capacity. The main focus of this project is to show some solutions for implementing a
communication system at the physical layer while keeping in mind the 5G requirements. A com-
munication system is composed of an efficient transmitter, that encodes messages into appropri-
ate waveforms (such as OFDM, GFDM, FBMC etc.) that can be conveyed through different types
of channels affected by different types of distortions such as fading and interference and then de-
coded correctly and efficiently back to the original data by a matched receiver.

OFDM vs GFDM: Spectral and Energy Efficiency

In 5G energy efficiency is very important because in certain scenarios millimeter waves are used,
which are very sensitive to environmental conditions such as obstacles, rain and other waves and
can only be used for transmission on relatively short distances. This is why in 5G cells are gener-
ally smaller. It is thus very important to ensure that the Power Spectral Density (PSD) of the side-
lobes in frequency domain of the encoded signal (OFDM or GFDM) are not causing further inter-
ference to adjacent transmitters. The subcarriers or frequencies available for transmission are
also limited and need to be reused within adjacent cells. GFDM can pack much more data into
larger blocks and is thus better suited for 5G systems. It has both better spectral efficiency and
better energy efficiency. OFDM has a better BER than GFDM in general but suffers from very high
out-of-band radiation. OFDM also has shorter blocks and wastes much more bandwidth for con-
trol information such as the CP. GFDM achieves a better PSD using adaptive filtering and employ-
ing one CP for much larger blocks. These conclusions apply both to when the CP is used and
when it is not.
 OFDM vs GFDM: Complexity
The complexity of OFDM transmitters and receivers is much lower than that of GFDM. From the
simulation it can be calculated that an OFDM encoding of 10 blocks of data is double as fast as a
GFDM encoding on average. However, it is double as slow to decode the same amount of data.
This result shows again that GFDM is better suited for 5G because faster decoding means lower
latency.
OFDM vs GFDM: Configurability
GFDM is very configurable. It is possible to use different filters at the transmitter and is possible
to use either the same filters or different at receiver according to the requirements of incumbent
applications or the type of subcarriers (each subcarrier suffers from different problems). This is
very important because one of the requirements of 5G is programmability according to the ser-
vice request. Several waveforms need to coexist and waveforms need to be considered holisti-
cally together with modulation schemes, hardware etc. GFDM allows for scenarios such as intro-
ducing sparsity at the signal encoding so that 6 users can be multiplexed together and served
with only 4 antennas.
References
http://www.sharetechnote.com/html/5G/5G_Phy_Candidate_GFDM.html
https://www.researchgate.net/post/Why-is-IFFT-block-used-in-OFDM-system-at-the-transmission-
side
https://dsp.stackexchange.com/questions/22473/what-is-difference-between-ofdm-and-gfdm
https://ieeexplore.ieee.org/abstract/document/6692619
https://dsp.stackexchange.com/questions/22473/what-is-difference-between-ofdm-and-gfdm