Introduction Implemenation Results Future Work Conclusions

GPU Based Polyphase Filter Banksfor VLBI

Mark McCurryChristopher Beaudoin & Geoff Crew

MIT Haystack REU Program 2011

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

Outline

IntroductionWhat is being done?PFBHardware

ImplemenationCUDA

ResultsQualityQuantity

Future Work

Conclusions

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

What is being done?

DSP

I Input of Analog values from data source near 4 Gb/s

I Perform Polyphase Filter Bank (PFB) on a Nvidia GPU

I Output of channelized frequency spectrum

I All of this done in realtime

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

What is being done?

DSP

I Input of Analog values from data source near 4 Gb/s

I Perform Polyphase Filter Bank (PFB) on a Nvidia GPU

I Output of channelized frequency spectrum

I All of this done in realtime

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

What is being done?

DSP

I Input of Analog values from data source near 4 Gb/s

I Perform Polyphase Filter Bank (PFB) on a Nvidia GPU

I Output of channelized frequency spectrum

I All of this done in realtime

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

What is being done?

DSP

I Input of Analog values from data source near 4 Gb/s

I Perform Polyphase Filter Bank (PFB) on a Nvidia GPU

I Output of channelized frequency spectrum

I All of this done in realtime

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

PFB

Polyphase Filter Bank

Two Channel Polyphase Filter Bank

x [n] r

z−1

↓2

↓2

E0(z)

E1(z)

FFT

f1[n]

f2[n]

b

b

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

Hardware

Roach Digital Backend (RDBE)

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

Hardware

Nvidia GPUs

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

CUDA

What is CUDA?

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

CUDA

CUDA

g l o b a l voidc u u n q u a n t i z e ( f l o a t ∗ dest , const i n t 8 t ∗ s r c ,

s i z e t N){

const s i z e t i = blockDim . x∗ b l o c k I d x . x +t h r e a d I d x . x ;

i f ( i <N)d e s t [ i ] = s t a t i c ca s t <f l oa t >( s r c [ i ] ) ;

}

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

CUDA

CUDA 2

// Setup v i a cudaMemcpy

// i npu t and output a r e on the GPUc u u n q u a n t i z e <<<10,10>>>(output , i n p u t , 1 0 0 ) ;

// Return v i a cudaMemcpy

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

Quality

Quality of output: Actual

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Frequency (half−cycles/second)

Nor

mal

ized

Gai

nMagnitude response of 256 tap FIR, 32 tap FFT based PFB

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

Quality

Quality of output: Matlab

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Frequency (half−cycles/second)

Nor

mal

ized

Gai

nMatlab Prototype PFB

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

Quantity

Speed

Table: Performance of GPU Code1

Performance Metrics Data Input Rate

Reference Implementaiton 744 MB/sNo extra channels 540 MB/sHardcoding FIR size 756 MB/sHardcoded FIR, Hand Tuned Block Size 890 MB/sUsing 1/10 cost 470GTX 637 MB/s

1As of August 1stGPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

Increasing Speeds

How to proceed:

I Profiling, and extensive testing

I Effective use of CUDA resources

I Waiting for new cards

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

Increasing Speeds

How to proceed:

I Profiling, and extensive testing

I Effective use of CUDA resources

I Waiting for new cards

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

Increasing Speeds

How to proceed:

I Profiling, and extensive testing

I Effective use of CUDA resources

I Waiting for new cards

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

Conclusions

I CUDA makes heavyweight software processing possible

I Real time processing is possible at lower rates

I GPU processing is applicable to Astronomy and scientificcomputation

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

Conclusions

I CUDA makes heavyweight software processing possible

I Real time processing is possible at lower rates

I GPU processing is applicable to Astronomy and scientificcomputation

GPU Based Polyphase Filter Banks for VLBI

Introduction Implemenation Results Future Work Conclusions

Conclusions

I CUDA makes heavyweight software processing possible

I Real time processing is possible at lower rates

I GPU processing is applicable to Astronomy and scientificcomputation

GPU Based Polyphase Filter Banks for VLBI