MultiRate 2

Download as pdf or txt
Download as pdf or txt
You are on page 1of 178

ELT-44007

Radio Architectures and Signal Processing


Review of Multirate Digital Signal Processing
Goals
• Understand the possibilities of basic multirate signal
processing techniques (sampling rate conversion by an
integer factor) in the context of advanced receiver (and
transmitter) architectures.
• Understand main principles for achieving
computationally efficient solutions for sampling rate
conversion by integer factor.
o Efficient multirate implementation structure for basic FIR
filters
o Efficiency of multistage designs.
o Polyphase structure, FIR and IIR Nth-band filters,
halfband filters as an important special case.
o Nyquist pulse shaping filtering (=separable Nth-band
designs) as an application of Nth-band filters
o CIC filter as an efficient multiplier-free structure for the
first stages of the decimation chain.
• Evaluating how the filter complexity depends on the
receiver (or transmitter) system design.
• Detailed studies on filter optimization techniques, alternative
filter structures, and extensive comparisons of different
solutions can be studied in the course SGN-23006 Advanced
Filter Design lectured by prof. Tapio Saramäki. Lecture
notes can be found at http://www.cs.tut.fi/~ts/.
• A highly (non-uniformly) decimated version of the lecture
notes of that course are used in this lecture. (Thanks, Tap!)
A practical guide for reading the multirate filter section
Main points:
• General idea of decimation and interpolation: pp. 5-43
• Efficient basic FIR structures: pp. 50-57
• “Noble identity”: pp.61-62
• Polyphase structures: pp. 63-71 or Invocom Course 4, Lect. 2
• Multistage designs: pp. 73-74, FIR decimator examples
• Idea of Nth-band FIR filter: pp.102-105
• Multistage Nth-band FIR filter, connection to Nyquist pulse
shaping principle: p. 113
• Half-band FIR filter: pp. 118-121
• CIC filter: pp. 157-167
Also quite useful:
• Half-band IIR filter: pp. 122-129, 146-155
• Idea of Nth-band IIR filter: pp. 130-137
What else:
• Numerous examples in Tapio’s slides give insight to the
characteristics of different design alternatives.
• Literacy of the strict formal representation style for design
specifications, etc., is required for deeper studies in multirate
signal processing.
• Why multirate signal processing is s good idea in
advanced transceiver architectures?
A general approach to increase the flexibility of receiver
implementations is to use wideband sampling in the receiver and
select the desired channel among the many digitized ones using
digital filtering, using a filter optimized for the particular
transmission system in use.
o In case of wideband sampling, the initial sampling rate is
much higher than symbol/chip rate used in baseband
processing.
The commonly used delta-sigma AD-conversion principle is also
based on heavy oversampling, even in the case of narrowband
sampling.
Concerning the DSP implementation complexity and power
consumption, it is very crucial to use the lowest possible sampling
rate at each stage of the processing chain (good examples of this
will be given later during the course).
o As a rule of thumb, for given (narrowband) selectivity
requirements and given input sampling rate, the
computational complexity (and power consumption) is, in a
well-designed multirate system, directly proportional to the
output sampling rate.
o Now think about the case where you could reduce the
sampling rate by a factor of 300 (could really be the case
in a wideband sampling receiver).
Similar ideas can be used also in the transmitter case:
synthesizing a high-rate, possibly multi-channel (also the term
multicarrier is used in this context) signal using DSP would
greatly improve the flexibility.
Later extensions within this course to
multirate signal processing tools

o Complex (I/Q) signals


o Bandpass signals
o Sampling rate conversion by a non-integer factor
o Low-order polynomial modelling in discrete-time for
modelling continuous-time signals or operators.
=> Reduction in implementation complexity, especially for
small N!
About Multi-Stage Nth-band Filters (added by MR)

Consider a multistage FIR decimator (or interpolator), where the


overall sampling rate conversion factor is partitioned as
N=N1·N2· … ·NK
and each of the stages is an Nk-th-band filter. It can be proven that
the cascade is an N-th-band filter!

This makes it possible to combine the benefits of multistage


decimators/interpolators and Nth-band FIR filters. One example
of a two-stage design is shown on the following pages.

A separable Nth-band filter is a cascade of two filters which have the


same amplitude responses and complementary phase responses, such
that the overall system has Nth-band characteristics and a linear phase
response.

Nth-band FIR filters and Nyquist pulse shaping filters are closely
related: Nth-band filter has zero intersymbol interference, i.e., it
satisfies the Nyquist pulse shaping criteria, both in time and frequency
domains.
• Raised-cosine filter is an Nth-band filter.
• Cascade of two root-raised-cosine filters is a special case of
separable Nth-band filters, where both stages have linear phase.
• Nth-band filter design methods can be used for optimizing pulse
shaping filters.
• Later, an example of a two-stage separable Nth-band filter is
shown. In this design, the stages have non-symmetric impulse
responses, but the cascade impulse response is symmetric.
o This is a matched filter pair and could, in principle, be used as RX
and TX filters in a communication system.
o Such a design may result on lower filter orders for given frequency
domain specifications.
o However, non-symmetric transmit filter impulse response
usually increases considerably the peak-to-average power ratio
of the transmitted signal, and is thus not useful in practice.
(M+1)/4 multiplications per input sample needed for
implementing a filter of length 2M+1.

=> A cascade of half-band filters is often a very efficient


choice for multirate signal processing!
Design of a Classical Lowpass and/or Highpass Filter
as a Parallel Connection of Two Allpass Filters

• Any odd-order classical lowpass or highpass IIR filter


(Butterworth, Chebyshev, inverse Chebyshev, elliptic) can
be implemented as a parallel connection of two allpass
filters. Also bandpass and bandstop filters can be
designed, with certain other restrictions on the filter order.
• This was first discovered in the case of wave digital lattice
filters, in which case the allpass filters are implemented
with special structures.

Design procedure using any filter package for classical


lowpass/highpass IIR filters:
1. Design a classical odd-order IIR filter of order 2N+1.
2. Include in A1(z) the outermost pole pair, the third outermost
pole pair, the fifth outermost pole pair, and so on. The
other pole pairs belong to A2(z). In the last step a single
pole, instead of a pole pair, is included in one of the
allpass filters.
3. Select the structures of the allpass filters and compute the
allpass filter coefficients. First- and second-order allpass
filters are commonly used as building blocks:
a + z −1 a2 + a1 z −1 + z −2
A ( z) =
(1)
, A ( z) =
(2)

1 + az −1 1 + a1 z −1 + a2 z −2

Alternating distribution of poles


to the allpass subfilters:
A complementary filter pair can be implemented (essentially)
with the complexity of a single filter!
Special Case: Halfband IIR Filters

• In this case, all the poles are on the imaginary axis, and
the first-order section reduces to a pure delay. The
second-order allpass sections are now of the form
a2 + z −2
A ( z) =
(2)
HB
1 + a2 z −2
This can be implemented using a first-order allpass section
where each unit delay has been replaced by two delays.

• Now only N coefficients are needed to implement a filter of


order 2N+1. => Very efficient!

Properties of halfband IIR filters:


• Transition band is symmetric around quarter of the
sampling rate.
• Lowpass and highpass filters have the same stopband
attenuations and pass band ripples. The passband and
stopband ripples are related through the power
complementarity property
• For any reasonable stopband attenuation is very small.

Halfband IIR filter is a special case of N-th-band IIR filters,


which will be discussed next. The connection to classical
filter theory is clear only in the half-band case.
Special Case 2: Hilbert Transform

• We can start with a lowpass IIR filter which can be


implemented as a parallel connection of two allpass filters.
• Then we introduce a frequency translation by π (i.e.,
quarter of the sampling rate):

• The frequency translation is achieved by replacing each


delay z-1 by jz-1.
• In the general lowpass case, the filter becomes complex.
• Using a halfband IIR filter, the transfer function takes a
special form and the implementation becomes simple
(ignoring the scaling factor 0.5 for simplicity)
H HT ( z ) = A1 (− z 2 ) + jz −1 A2 (− z 2 ) I
A1(-z2)

z-1A2(-z2) Q

• The phase difference of the two allpass filters is


approximately π/2 outside the transition band regions.
Increasing the stopband attenuation improves the phase
approximation.
• Using approximately linear phase halfband IIR filter (see
later) results in a Hilbert transformer in the usual sense.

You might also like