FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier...
Transcript of FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier...
![Page 1: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/1.jpg)
FFT Algorithms and Architectures
MOJTABA MAHDAVI
![Page 2: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/2.jpg)
Outline
Discrete Fourier Transform (DFT)
Fast Fourier Transform (FFT)
Twiddle Factor Multiplication
FFT Algorithms
FFT Architectures
Data Flow Processing
DIF vs. DIT Decomposition
![Page 3: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/3.jpg)
Outline
Discrete Fourier Transform (DFT)
Fast Fourier Transform (FFT)
Twiddle Factor Multiplication
FFT Algorithms
FFT Architectures
Data Flow Processing
DIF vs. DIT Decomposition
![Page 4: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/4.jpg)
Discrete Fourier Transform (DFT)
DFT is one of the most important algorithms in Digital
Signal Processing (DSP).
DFT is widely used in several applications:
o Audio and Image Processing
o Spectrum Analysis of Signals
o Digital Communication Transmitter/Receivers
![Page 5: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/5.jpg)
Spectrum Analysis
DFT calculates the frequency spectrum of a signal
(discrete sinusoids components) to examine the
information encoded in:
o Frequency
o Phase
o Amplitude
DFT can find a system's frequency response from the
system's impulse response and vice versa
o Analyze the frequency/time-domain behavior of a system
![Page 6: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/6.jpg)
Spectrum Analysis
![Page 7: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/7.jpg)
Digital Communication Transmitter/Receiver
DFT is extensively used in multi-carrier transmission
systems like orthogonal frequency domain multiplexing
(OFDM).
DFT and IDFT are used to perform OFDM demodulation
and modulation, respectively.
![Page 8: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/8.jpg)
Digital Communication Transmitter/Receiver
Simple OFDM physical layer chain.
![Page 9: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/9.jpg)
DFT
An N-point DFT is calculated as:
Twiddle factors:
Complexity:
![Page 10: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/10.jpg)
8-point DFT
![Page 11: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/11.jpg)
Outline
Discrete Fourier Transform (DFT)
Fast Fourier Transform (FFT)
Twiddle Factor Multiplication
FFT Algorithms
FFT Architectures
Examples
DIF vs. DIT Decomposition
![Page 12: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/12.jpg)
Fast Fourier Transform (FFT)
Several fast Fourier transform algorithms have been
proposed to reduce the computational complexity of DFT
calculation:
o Prime factor algorithm
o Winograd algorithm
o Cooley-Tukey algorithm
• Most common
• Focus of this presentation
![Page 13: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/13.jpg)
Fast Fourier Transform (FFT)
FFT employs the symmetry and periodic properties of the
twiddle factors:
, …..
FFT reduces the computational complexity of DFT
calculation to:
![Page 14: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/14.jpg)
Complexity Reduction
![Page 15: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/15.jpg)
Outline
Discrete Fourier Transform (DFT)
Fast Fourier Transform (FFT)
Twiddle Factor Multiplication
FFT Algorithms
FFT Architectures
Data Flow Processing
DIF vs. DIT Decomposition
![Page 16: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/16.jpg)
Twiddle Factor Multiplication
There are two types of Twiddle factor multiplications:
o Trivial
• Multiplication by±1,±j
• Rotation, …
o Non-trivial
• Complex Multiplications
![Page 17: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/17.jpg)
Twiddle Factor Multiplication
N=4 N=8 N=16
j
1
-j
-1
j
1
-j
-1
j
1
-j
-1
Trivial Non-Trivial
![Page 18: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/18.jpg)
Trivial Rotation
Trivial rotation can be realized by:
o Interchanging the real and imaginary parts and/or
o Changing the sign of the real and/or imaginary parts of the
input data
![Page 19: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/19.jpg)
Non Trivial Rotation
Non trivial rotation can be implemented using:
o General complex multiplier
• To perform any non-trivial multiplication
o Constant multiplier
• To perform non-trivial multiplications for specific coefficients
• Less area
o CORDIC algorithm
• To realize the non-trivial multiplications through rotation
![Page 20: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/20.jpg)
4-point DFT
)3(
)2(
)1(
)0(
)3(
)2(
)1(
)0(
4/184/124/60
4/124/84/40
4/64/44/20
0000
x
x
x
x
eeee
eeee
eeee
eeee
X
X
X
X
jjj
jjj
jjj
)3(
)2(
)1(
)0(
11
1111
11
1111
)3(
)2(
)1(
)0(
x
x
x
x
jj
jj
X
X
X
X
Only trivial coefficients
j
1
-j
-1
![Page 21: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/21.jpg)
Add
ROM
CA
EG
Region Det.A 𝑪𝒐𝒔 𝜶 − 𝒋𝑺𝒊𝒏(𝜶)
B 𝑺𝒊𝒏 𝜶 − 𝒋𝑪𝒐𝒔(𝜶) C −𝑺𝒊𝒏 𝜶 − 𝒋𝑪𝒐𝒔(𝜶) D −𝑪𝒐𝒔 𝜶 − 𝒋𝑺𝒊𝒏(𝜶) E −𝑪𝒐𝒔 𝜶 + 𝒋𝑺𝒊𝒏(𝜶) F −𝑺𝒊𝒏 𝜶 + 𝒋𝑪𝒐𝒔(𝜶) G 𝑺𝒊𝒏 𝜶 + 𝒋𝑪𝒐𝒔(𝜶) H 𝑪𝒐𝒔 𝜶 + 𝒋𝑺𝒊𝒏(𝜶)
Im{input}Re{}
Im{}
Re{input}
General Twiddle Factor Multiplier
ROM size Reduction:o Based on the symmetry property, only the coefficients in the first Π/4
region are saved in ROM
o Mapping Table will extract the other coefficients j
1
-j
-1
Concept: Using Coefficient Symmetry
![Page 22: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/22.jpg)
Twiddle Factor Multiplication– Constant Multiplier
>> 2
>> 2
>> 4
>> 2
>> 2 >> 4
1 2 − 𝑗 1 2 ∗ 𝑎 + 𝑗𝑏 = 𝑐 + 𝑗𝑑
𝑏
𝑐
𝑎
𝑑
j
1
-j
-1
Concept: CSD Representation
![Page 23: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/23.jpg)
Outline
Discrete Fourier Transform (DFT)
Fast Fourier Transform (FFT)
Twiddle Factor Multiplication
FFT Algorithms
FFT Architectures
Data Flow Processing
DIF vs. DIT Decomposition
![Page 24: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/24.jpg)
FFT Algorithm
The most popular FFT algorithms are:
o Radix-r
o Improved FFT (Radix-2𝑛)
o Mixed-radix
o Split-radix
![Page 25: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/25.jpg)
Radix-r Algorithm
The radix-r FFT algorithms:
o DFT of length N is recursively decomposed into N/r and runtil all the remaining transform lengths are less than or
equal to r.
o Number of stages: 𝐥𝐨𝐠𝒓 𝑵
o A high radix FFT algorithm reduces the number of
processing stages
• Increases the hardware complexity of each stage significantly.
![Page 26: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/26.jpg)
Radix-2 Butterfly
Complex inputs/outputs
![Page 27: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/27.jpg)
Radix-4 Butterfly
![Page 28: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/28.jpg)
4-point FFT with Radix-2 Butterfly
x(0)
x(1)
x(2)
x(3)-j
X(0)
X(2)
X(1)
X(3)
![Page 29: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/29.jpg)
16-point FFT - Radix-2 Algorithm
)(log 2 NNumber of stages:
Stage 1 Stage 2 Stage 3 Stage 4
Twiddle Factor Multiplications
Many of the coefficients are trivial:
1, -1, j, -j
![Page 30: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/30.jpg)
16-point Radix-4 FFT
Stage 1 Stage 2
![Page 31: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/31.jpg)
Small-radix vs. High-radix FFT Algorithm
Selection of radix has a large impact on the complexity of
FFT algorithm
Small radix FFT architecture:
Simple butterfly operation
Higher number of twiddle factor multiplications
High-radix pipelined FFT architectures have been proposed
to improve the arithmetic resource utilization.
![Page 32: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/32.jpg)
Small-radix vs. High-radix FFT Algorithm
High-radix FFT:
o The more efficient use of multipliers and adders
o Less number of twiddle factor multiplications
o Reduces the number of stages
o More complexity in trivial twiddle factor computation
o More complex stage (i.e. butterfly units)
• Radices higher than 4 require butterflies with non trivial rotations.
![Page 33: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/33.jpg)
Improved FFT (Radix-2^n) Algorithm
Radix- 2𝑛 algorithms are proposed to overcome the
drawback of high-radix algorithms.
Radix-2𝑛 algorithm can be explained by applying the CT
algorithm two times.
o Basic unit of decomposition consists of the radix-2 butterfly.
o The number of stages requiring twiddle factor multiplications
is reduced.
![Page 34: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/34.jpg)
e.g. Radix-2^2 Algorithm
This algorithm has:
o The same number of non-trivial multiplications as a radix-4
algorithm
o The same butterfly structure as that of radix-2 algorithm
• Can be mapped to radix-2 butterflies.
This can further on extended to Radix-2^3 and 2^4.
![Page 35: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/35.jpg)
Mixed-radix Algorithm
The mixed-radix algorithms can be derived by mixing
different radixes.
o Generate desired FFT lengths
o More efficient processing
o Hardware complexity is similar to radix-2^n algorithm
![Page 36: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/36.jpg)
Split-radix Algorithm
The main idea is that independent parts of the algorithm
should be computed independently based on the best
possible computational scheme.
o Reduction in computational complexity
![Page 37: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/37.jpg)
Split-radix Algorithm
In split-radix algorithms for 2𝑛 size DFT:
o The total number of complex multiplications can be reduced
o Each stage becomes irregular
o Not efficient in terms of pipelined processing
o More complex control due to the irregularity
![Page 38: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/38.jpg)
Outline
Discrete Fourier Transform (DFT)
Fast Fourier Transform (FFT)
Twiddle Factor Multiplication
FFT Algorithms
FFT Architectures
Data Flow Processing
DIF vs. DIT Decomposition
![Page 39: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/39.jpg)
FFT Architectures
Most FFT architectures can be categorized into:
o Direct implementation
o Memory-based
o Pipelined
![Page 40: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/40.jpg)
Direct Implementation
Requires a number of processing elements equal to the
number of operations
o Very hardware intensive
o It can be suitable for small size FFTs
o The utilization of the butterflies and rotators is 100%
![Page 41: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/41.jpg)
Memory Based Architectures
One or more processing elements (Pes) calculate all the
butterflies and twiddle factor multiplications.
o It is necessary to compute whole FFT before it receives new
samples.
o Unable to compute the FFT when data arrives continuously.
• This can be solved by adding extra memory
Concept: Folding & Time Multiplexing
![Page 42: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/42.jpg)
Memory Based Architectures
Concept: Unfolding/Parallel Processing
![Page 43: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/43.jpg)
Memory Based Architectures
Memory-based architectures (in-place architecture):
o Smaller area
o Low power
o Long latency
o Require additional buffer space
o Lower throughput compared to the pipelined architectures
• Parallel processing is used to improve throughput and latency.
– Hardware cost is increased
• High-radix processing elements are used to improve throughput.
– It causes memory conflict problems
Not suitable for FFT computation in real time applications
![Page 44: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/44.jpg)
Pipelined Architectures
Two principal techniques for pipelined architectures:
o Delay Feedback (DF), often referred to as Feedback
• SDF
• MDF
o Delay Commutator (DC), often referred to as Feed Forward (FF)
• MDC
• SDC
Pipelined architecture is a proper choice for high-
throughput and real time applications
![Page 45: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/45.jpg)
Single-path Delay Feedback Architectures
SDF-based architectures provide memory feedback paths
to manage some butterfly outputs during each stage.
SDF techniques allow the initial FFT output sample to be
generated instantly after the final FFT input sample has
been processed.
SDF architecture has one continuous data stream of one
sample per clock cycle
![Page 46: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/46.jpg)
Pipeline Architectures
The pipelined FFT architectures:
o Higher throughput
o Lower latency
o Suitable for real-time applications
o Acceptable hardware cost
o Perform non-stop processing at sample rate
o Proper for low power solution
![Page 47: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/47.jpg)
Single-path Delay Feedback Architectures
![Page 48: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/48.jpg)
Single-path Delay Feedback Architectures
SDF architecture has:
o Lower Latency!
o Low cost
o High hardware efficiency
o Low throughput due to the single path
• No concurrent processing
o Arithmetic utilization is relatively low (50%)
SDF is an optimal choice in terms of the hardware
cost and performance for many applications
![Page 49: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/49.jpg)
Multipath Delay Feedback Architectures
MDF architecture can be generated by extending the SDF
FFT architecture using a multiple-path approach.
o A solution to provide a higher throughput
o Higher hardware cost
o Arithmetic utilization is relatively low (50%)
Multiple-path (M) architectures, are often adopted for high
throughput applications
![Page 50: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/50.jpg)
Multipath Delay Feedback Architectures
Concept: Unfolding/Parallel Processing
![Page 51: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/51.jpg)
Multi Delay Commutator Architectures
MDC-based architectures replace feedback data paths
with feed forward data paths with commutators as
switching operations.
o Each stage forwards its output to the next without any
feedback
o MDC architecture processes several samples in parallel
These architectures can be improved by using radix-2^n.
![Page 52: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/52.jpg)
Multi Delay Commutator Architectures
![Page 53: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/53.jpg)
Multi Delay Commutator Architectures
MDC-based architecture:
o Simple control path
o 100% utilization ratio of butterflies
o Higher throughput than SDF
o Higher hardware cost
MDC can achieve higher throughput, while
SDF needs less memory and hardware cost.
![Page 54: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/54.jpg)
Algorithm/Architecture Comparison
![Page 55: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/55.jpg)
Outline
Discrete Fourier Transform (DFT)
Fast Fourier Transform (FFT)
Twiddle Factor Multiplication
FFT Algorithms
FFT Architectures
Data Flow Processing
DIF vs. DIT Decomposition
![Page 56: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/56.jpg)
SDF Architecture
16-point Single-input Pipelined FFT
XkRadix2
ButterflyXin
Mem 18 Words
Radix2 Butterfly
Mem 24 Words
Radix2 Butterfly
Mem 32 Words
Radix2 Butterfly
Mem 41 Word
Stage 1 Stage 2 Stage 3 Stage 4
![Page 57: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/57.jpg)
16-point SDF Architexture
![Page 58: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/58.jpg)
Timing Diagram of Pipelined FFT
Latency
![Page 59: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/59.jpg)
2048-point SDF Architecture
Symbol 1
2048 Samples
Symbol 2
2048 Samples
Symbol 1
2048 Samples
Symbol 2
2048 Samples
Symbol 3
2048 Samples
IFFT In:
IFFT Out:Latency=2048 CC
![Page 60: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/60.jpg)
IFFT
IFFT is realized as:
The same hardware can be used
![Page 61: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/61.jpg)
Outline
Discrete Fourier Transform (DFT)
Fast Fourier Transform (FFT)
Twiddle Factor Multiplication
FFT Algorithms
FFT Architectures
Data Flow Processing
DIF vs. DIT Decomposition
![Page 62: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/62.jpg)
4-point FFT with Radix-2 Butterfly
x(0)
x(2)
x(1)
x(3)-j
X(0)
X(1)
X(2)
X(3)
![Page 63: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/63.jpg)
4-point FFT with Radix-2 Butterfly
x(0)
x(1)
x(2)
x(3)-j
X(0)
X(2)
X(1)
X(3)
![Page 64: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/64.jpg)
DIF vs. DIT Decomposition
According to the decomposition direction, FFT algorithms
can be classified into:
o DIF decomposition:
• The output sequence is separated into even and odd indexed
samples iteratively.
o DIT decomposition:
• Separates the input sequence into even and odd samples
iteratively.
![Page 65: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/65.jpg)
DIF vs. DIT Decomposition
In DIF, the input samples are usually in order and the
output samples are in bit-reversed order.
In DIT, the input samples are usually in bit-reversed order
and the output samples are in natural order.
• The location of the twiddle factor multiplications
• Input/Output Order
![Page 66: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/66.jpg)
16-point DIF
![Page 67: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/67.jpg)
16-point DIT
![Page 68: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/68.jpg)
16-Point FFT
X(3) - 0011 X(12) - 1100
X(15) - 1111 X(15) - 1111
X(7) - 0111 X(14) - 1110
X(11) - 1011 X(13) - 1101
X(1) - 0001 X(8) - 1000
X(13) - 1101 X(11) - 1011
X(5) - 0101 X(10) - 1010
X(9) - 1001 X(9) - 1001
x(12)
x(15)
x(14)
x(13)
x(8)
x(11)
x(10)
x(9)
W 8
X(2) - 0010 X(4) - 0100
X(14) - 1110 X(7) - 0111
X(6) - 0110 X(6) - 0110
X(10) - 1010 X(5) - 0101W 2
W 6
X(0)
X(12) - 1100 X(3) - 0011
X(4) - 0100 X(2) - 0010
X(8) - 1000 X(1) - 0001
x(4)
x(7)
x(6)
x(5)
x(0)
x(3)
x(2)
x(1)
W 7
W 5
W 6
W 4
W 3
W 1
W 2
W 0
W 6
W 2
W 4
W 0
W 0
W 0
W 4
W 0
W 4
W 0
W 4
W 0
W 4
Bit-reversed ”Normal”
A Reordering Circuit is needed to perform the above conversion
![Page 69: FFT Algorithms and ArchitecturesFFT Architectures Examples DIF vs. DIT Decomposition. Fast Fourier Transform (FFT) Several fast Fourier transform algorithms have been proposed to reduce](https://reader030.fdocuments.net/reader030/viewer/2022012008/611d36493ffea8135e56d437/html5/thumbnails/69.jpg)