Polyphase Channelizer

Transformation

1/259
 Transformation (Step 1)

Equivalence Theorem:

”Operation of down-conversion, followed by a LPF are totally equivalent to the

operation of BPF followed by a down conversion”

y ( n , k ) = [ x ( n ) e − jn θ k ] * h ( n )
N −1
y (n, k ) = ∑ x [ n − r ]e − j θ k
( n−r )
h(r )
r =0
N −1 jr θ k
= ∑ x [ n − r ]e
r =0
− jn θ k
h (r )e
N −1 jr θ k
=e − jn θ k
∑ x[ n − r ]h ( r ) e
r =0

2/259
Equivalence Theorem &
Sequence of maneuvers

• Slide the input heterodyne through the lowpass

filter to their output
• By doing so, it converts the lowpass filter to a
complex bandpass filter
• Slide the output heterodyne to the downside of
the down-sampler
• Doing so, it aliases the centre frequency of the
oscillator
• Restrict the centre frequency of bandpass to be
a multiple of the output sample rate.
• Doing so, assure aliases of the selected
passband to the baseband by re-sampling
operation
• Discard the un-necessary heterodyne

Note: θk should be an integer multiple of (2π/M)

3/259
 Transformation (Step 2)
Noble Identity:

” A filter processing every Mth input sample followed by an output M-

to-1 down sampler is the same as an input M-to-1 down sampler
followed by a filter processing every Mth input sample”

4/259
Transformation (Step 2)

N −1
H (z) = ∑
n=0
Z −n
h(n)
M −1
= ∑
r =0
Z −r
H r (Z M
)
M −1 ( N / M ) −1
= ∑Z
r =0
−r
∑ h ( r + nM
n=0
)Z − Mn

5/259
 Transformation (Step 2)

N −1
H ( z ) = ∑ Z − h(n)
n

n=0

G ( z ) = H ( z ) | z = e jϑ z = H ( e − j ϑ z )
M −1
H ( ze − j ( 2π / M ) k
)= ∑ e
Z − r j ( 2 π / M ) rk

r =0
H r ( z)
M −1
y ( nM , k ) = ∑
r =0
y r ( nM ) e j ( 2π / M ) rk

fs = N * Δ f

Re-sampling M-Path down converter

6/259
Phase Coherent summation & FFT

7/259
 Polyphase Channelizer

• Commutator to down-sample the data rate

• Polyphase partitioned filter
• Complex phase rotators to extract the
individual channels
• Polyphase parameters
– Input sampling frequency (fs)
• Inter-carrier spacing (∆f)
• Number of channels (M)

8/259
 Pipelined Frequency Transform

• Based on binary tree of down-converter and

sample-rate converter
• Divide the input band into two halves with half
sampling rate
• Again split each half band into two sub-bands, and
so on un-till the last tree level produces the
required seperated channels
• More expensive in term of silicon area, because of
many more single channel channelizer
requirements
• Less flexible as it require channels to be equal
bandwidth and uniformly distributed.

9/259
 Comparison & Selection

Per-channel approach wins in many aspects, but its implementation for

high number of channels is infeasible
Polyphase channelizer is most suitable for SDR wideband channelizer
front end

10/259
 Polyphase filter bank parameters

Sampling Frequency, Number of Channels, Spectral Spacing,

Output Sample Rate.

fs = N * Δ f
• DFT performs the task of seperating the channels after polyphase
filter, so it is natural to conclude that transform size is equal to
number of channels.
• The filter bandwidth is determined by the weights of the lowpass
prototype filter, and it is common for all the channels.
• Channelizer is used to seperate the adjacent communication
channels, which are characterized by the specifc center frequency
and non-overlapping bandwidth

11/259
 Maximally & Non-Maximally
Decimated System

”A system is said to be Maximally Decimated when the ouput

sample rate is equal to the inter-channel spacing otherwise it is
non-maximally decimated system ”

Maximally Decimated

Non-Maximally Decimated

12/259
 Computational Complexity

Work Load = Filter Length / Sample Ratio

= (no.of ops/output) / (input/output)
= no.of ops / input

13/259
Prototype Filter Work Load

14/259
Polyphase Filter Work Load

15/259
 Computational Complexity

Does It make sense to use Polyphase filter for the

Channelization ...... ???

Yes

16/259
 Computational Complexity
1792 ops. once per 48 input
work load = 1792/48 = 38 ops/input.
Note:
Polyphase channelizer forms 64 channels at the output and therefore, we
conclude that polyphase form should be used even if just few output
channels are required.

17/259