FBMC-OQAM in Doubly-Selective Channels:

A New Perspective on MMSE Equalization

Realised by : Professor :
● NAHRI Mouad Mr ETTOLBA Mohamed
● LAARIF Amen Allah
Outline to adopt
● Introduction
● FBMC-OQAM
● Equalization
○ MMSE
○ Interference cancellation

● Extension to MIMO

● FBMC Application
● Conclusion
2
 1

Introduction
❖ OFDM : Orthogonal Frequency Division Multiplexing

A modulation technique that distributes the signal over a large number of individually modulated
orthogonal subcarriers at a low rate. In addition to a spectral efficiency very close to the optimum,
this technique provides excellent resistance to frequency fading and reduce interference.

4
❖ FBMC : Filter Bank Multi-Carrier
Filter Bank Multicarrier (FBMC) systems are a subclass of multicarrier (MC) systems. While its
basic principle, dividing frequency spectrum into many narrow subchannels. It assures:
•Good time-frequency localization
•Full spectral efficiency

5
❖ MSE : Mean Squared Error
The MSE assesses the quality of an estimator measures the average of the
squares of the errors that is, the average squared difference between the
estimated values and what is estimated.

6
❖ MMSE : Minimum Mean Squared Error
MMSE is an estimation method which minimizes the mean square error
(MSE), which is a common measure of estimator quality, of the fitted values of
a dependent variable.

7
 2

FBMC-OQAM
OFDM biggest disadvantages is the poor spectral properties.
Solution ⇒ FBMC (Filter Bank Multi-Carrier)
Cause it present a simple orthogonality more than OFDM with special treatment.

On a channel, where is Interference VS noise: interference can be neglected.

⇒ BER(OFDM) = BER(FBMC)
In some scenarios, when SNR increase, interference may increase too ,interference may be dominant,
so that sophisticated equalization methods become necessary.

9
FBMC assume a time-invariant channel:
Methode 1: propose a parallel equalization scheme, requiring multiple parallel Fast
Fourier Transform (FFT) blocks
Methode 2: propose a larger FFT

In this presentation, we consider an equalization method that operates after a conventional FFT.

Note:
In FBMC, real-valued data symbols are transmitted over a rectangular time-frequency grid.
denoting the transmitted data symbol at subcarrier position l and time-position k by

10
Optimal equalizer ⇒ consider the interference contribution of:

neighboring subcarriers neighboring time-symbols

= Transmitted data symbol => Signal transmitted : L subcarriers and K time-symbols.

Where:

Note: In FBMC, p(t) is complex orthogonal to TF = 2 ==> Rate increase

However we reduced TF to TF=1/2 in order to shift the interference caused

to the purely imaginary domain, due to this imaginary interference, only real-valued symbols can be
transmitted. ==> interference decrease

11
In discrete time-mode , (1) can be expressed by:

PS: s is generated by an IFFT and not by a matrix multiplication

In case:

We model time-variant multipath propagation by a time variant convolution matrix.

The overall transmission system can be expressed by:

Note that real orthogonality in FBMC implies that

12
 3

Equalization
In many practical scenarios, the off-diagonal elements of D, are so small
⇒ that they are dominated by the noise.

Solution : A simple one-tap Zero Forcing (ZF) equalizer.

representing the appropriate diagonal element of D

Adapt off diagonal terms to diagonal ones, is then sufficient to achieve a close to optimal
symbol detection probability.

For white Gaussian noise, a one-tap ZF represent a good equalizer but in highly selective
channel where operating a high SNR mode, one top ZF might not be efficient.

14
To better illustrate this fact, let us consider a one-tap equalizer in case of a Vehicular channel model
and a Jakes-Doppler spectrum with a maximum doppler shift of 1.16 kHz.

The study will be done on the realization of two types of Filter prototype: PHYDYAS vs HERMITE.

Note:

Prototype filters are models of electronic filters that make it possible to determine all filter benches.

15
As illustrated in Fig.1, only a few close symbols in time and frequency contribute
significantly to the SIR.
Similar, Fig.2 shows the interference distribution for the Hermite prototype filter. The total
SIR is 23 dB and thus better than for the PHYDYAS filter.
From both figures, we conclude that the underlying prototype filter has a huge impact on
which neighboring symbols cause interference.

16
Comparaison: PHYDYAS & HERMITE

PHYDYAS HERMITE

SIR 20 dB 23 dB

DISTRIBUTION LONGITUDINAL RADIAN

FREQUENCY LOCALISATION {-1,1} DIVERSITY

CONCENTRATION

SYMMETRIC INTERFERENCE
SYMMETRY CONTRIBUTION
(for Time & frequency)

When SIR > 10 dB , one TAP equalizer car be sufficient , in other cases , we should require
more equalisation methods.

17
We thus require more sophisticated equalization methods:

○ MMSE

○ Interference cancellation

18
● MMSE Equalization
A simple way to estimate the transmitted data symbol vector x would be the employment of a
simple conventional MMSE equalizer.
But , this does not work because such equalizer would also include the imaginary interference
→ large detection errors.

Solution : we stack real and imaginary part in a large vector.

The full block MMSE equalization of y, can then be calculated by:

19
There are 2 problems with getting the full block :
❏ It has a high complexity to define the inverse matrix.
❏ A large Delay is introduced by waiting all symbols to be received before launching equalizer
performing.
⇒ Solution : We adopt an n-tap MMSE equalizer which includes only a few neighboring symbols.
Where the transmission model of (6) according to is:

20
 is a subvector of y with S representing the considered subblock.
For example, a subblock vector of size |S| = 5 (5-tap) can be written as:

Such subblock was used in (10) includes neighboring symbols in time but ignores neighboring subcarriers
⇒ it is not optimal.

21
So the |S|-tap MMSE equalization of (10) becomes :

we require in total LK matrix inversions of size 2|S| × 2|S|.

⇒ The computational complexity is usually lower then for the full block MMSE equalizer but still
relatively high.

22
The n-tap equalizer utilizes n received symbols to estimate the transmitted data symbol at
one time-frequency position (indicated by the crosses in the center)

23
The figure shows how the BER depends on the SNR for the PHYDYAS prototype filter.
MMSE equalization will result in a performance between these two reference curves (One-Tap,
ZF & Doubly-Flat)

Even the 13-tap equalizer already achieves a performance close to that of the full block MMSE equalizer.

24
So the Hermite prototype filter performs better than the PHYDYAS prototype filter in high velocity
scenarios, we consider only the Hermite prototype filter for the next part.

25
● Interference cancellation
The MMSE equalizer has the big disadvantage of high computational complexity, even if we
consider only a small subblock.
We therefore propose a simple, yet effective, interference cancellation scheme.
Similar as suggested for OFDM, we can cancel the interference by:

Thus (D− diag{D}) represents all the off-diagonal elements of D. The data estimates are
obtained by one-tap equalization according to:

26
The BER of interference cancellation lies between the 5-tap equalizer and the 9-tap equalizer. After
2 to 3 iterations, we no longer see a significant performance improvement. Thus, more than 2 to 3
iterations are not necessary. Although the cancellation scheme performs not as good as the MMSE
equalizer in terms of BER, the performance is still very good, especially for practical relevant SNR
values smaller than 30 dB.

27
 4

Extension to MIMO
So far, we considered only Single-Input and Single-Output (SISO) transmissions, but the
extension to MIMO is straightforward thanks to our matrix notation.
We only need bigger matrices. Of course, this further increases the computational complexity,
making its practical implementation challenging.

29
As a theoretical reference, we consider 2 × 2 MIMO transmission system, which can be
modeled by :

Similar to (8), we find the full block MMSE equalizer by:

with noise matrix

30
In a similar way, we can also find the subblock MIMO MMSE equalizer.
We observe a similar behavior as in SISO. Considering only neighboring time-symbols, but ignoring
neighboring subcarriers, is highly suboptimal. Thus, our 9-tap equalizer performs much better than the
9-tap equalizer proposed in (10)

31
 5

FBMC Application
Multi-carrier filter-based modulations (FBMC) are among the most promising candidates for the future 5G.

Compared to the state of the art based on the OFDM modulation used in 4G, the FBMC coupled to the
OQAM (Offset QAM) has proven advantages in terms of :

• spectral efficiency,
• shape of the spectrum,
• Resistance to synchronization errors and Doppler effect.

These advantages are key elements to respond to the "5G challenges" resulting from the extension of the
use of current networks to new scenarios and use cases (eg moving networks, massive-machine
communications, ultra dense networks, etc.) .

33
 6

Conclusion
Considering only neighboring time-symbols for the n-tap, MMSE equalizer, as usually done in
literature for FBMC, is not optimal, especially for the Hermite prototype filter and ina highly
time-variant channel.

The MMSE equalizer has a high computational complexity, so that it is questionable whether
it can be employed in practical systems.

As a theoretical reference, however, it is quite useful. A simple interference cancellation

scheme can solve the complexity issue and achieves a good performance for practical
relevant SNR ranges.

35
Thanks!
Any questions?

36