Y. C. Jenq1 Non-uniform Sampling Signals and Systems (A/D & D/A Converters) Y. C. Jenq Department of...

Post on 22-Dec-2015

217 views 2 download

transcript

Y. C. Jenq 1

Non-uniform Sampling Signals and Systems

(A/D & D/A Converters)

Department of Electrical & Computer Engineering

Portland State University

P. O. Box 751

Portland, OR 97207

jenq@ece.pdx.edu

Y. C. Jenq 2

Outlines

Non-uniform Sampling Signals Digital Spectrum of Non-uniformly

Sampled Signal Timing Error Estimation Reconstruction of Digital Spectrum

Y. C. Jenq 3

Non-uniform Sampling

time, t

Waveform amplitude, x(t) with FT = Xc()

t0 t1 t5t4t3t2 t7t6 t8

T = nominal sampling periodn = tn- nT, rn = n / T

T

1 2

M=4

Y. C. Jenq 4

Non-uniform Sampling Clock

t0 t1 t5t4t3t2 t7t6 t8

T

Y. C. Jenq 5

Non-uniform Sampling Examples

Random Equivalent–time Sampling Interleaved ADC Array Direct Digital Synthesizer

Y. C. Jenq 6

Random Equivalent-Time Sampling

Triggering LevelTriggering Time Instances

Sampling Time Instances

Y. C. Jenq 7

Random Equivalent-Time Sampling

Y. C. Jenq 8

Interleaved ADC Arrays

ADC

ADC

ADC

ADC

Signal in

Delay elements

Memory

Memory

Memory

MemoryOR with a 4-phase clock

Sampling Clock

Y. C. Jenq 9

Interleaved ADC ArraysADC

ADC

ADC

ADC

Signal in

Memory

Memory

Memory

Memory

4-phase clock

Y. C. Jenq 10

Direct Digital Synthesizer (DDS)

WaveformMemory

WaveformMemory

Phase AccumulatorPhase Accumulator

D/A Converter

D/A Converter

Low-Pass Filter

Low-Pass Filter

Y. C. Jenq 11

Direct Digital Synthesizer (DDS)

WaveformMemory

WaveformMemory

D/A Converter

D/A Converter

Low-Pass Filter

Low-Pass Filter

Integer Part Fraction

Integer Part Fraction

+

Address Accumulator

Address Increment Register

Y. C. Jenq 12

Direct Digital Synthesizer (DDS)Waveform Memory

Fs: Master Clock Frequencyf: Sine Wave FrequencyTL: Table Length

Y. C. Jenq 13

Direct Digital Synthesizer (DDS)Frequency Resolution

Integer Part Fraction

W + L/M

B bits

Frequency Resolution = Fs/2B-1

Sine wave Frequency f = (W+L/M)Fs/TL

Y. C. Jenq 14

Non-uniform Sampling Model

T = nominal sampling period tn = nT + n , and n is periodic with period M.

Let n = k M + m where k ranges from –∞ to +∞ and m ranges from 0 to (M-1), Then

tn = ( k M + m )T + (kM+m)

= k M T + m T + m

= k M T + m T + rm T

where rm = m/T

Y. C. Jenq 15

Yih-Chyun Jenq, “Digital Spectra of Non-uniformly Sampled Signals - Fundamentals and High-Speed Waveform Digitizers,” IEEE Transactions on Instrumentation and Measurement, vol. 37, no. 2, June 1988.

Yih-Chyun Jenq, “Digital Spectra of Non-uniformly Sampled Signals: A Robust Time Offset Estimation Algorithm for Ultra High-Speed Waveform Digitizers Using Interleaving,” IEEE Transactions on Instrumentation and Measurement, vol. 39, no. 1, February 1990

Digital Spectrum of Non-uniformly Sampled Signals

Y. C. Jenq 16

Digital Spectrum of Non-uniformly Sampled SignalsIf we use x(tn) to compute the digital spectrum, Xd(), as if the data points were sampled uniformly, i.e.,

Xd() = n x(tn) e-jn

Then, it can be shown that

Xd(T) = (1/T)k A(k,) Xc[-k(2/MT)]Where

A(k,) = (1/M)m=0,(M-1) e-j[-k(2/MT]rmTe-jkm(2/M)

Notice that A(k,) is the m-point DFT of e-j[-k(2/MT]rmT

Y. C. Jenq 17

Digital Spectrum of Non-uniformly Sampled Sinusoid

Input Signal x(t) = exp(jot), And Xc()=2()

Then Xd() = (2/T) k A(k) [-o-k(2/MT)]

where A(k) =m=0,(M-1)(1/M)ejrmoTe-jkm(2/M)

Notice that A(k) is no longer a function of

and A(k) is a M-point DFT of ejrmoT, m=0, 1,…,M-1

Y. C. Jenq 18

Digital Spectrum of Non-uniformly Sampled Sinusoid

A(0)

A(1)A(2)

A(3)

M=4

Y. C. Jenq 19

Digital Spectrum of Non-uniformly Sampled Sinusoid

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-20

-10

0

10

20

30

40

50

60Spectrum of Non-uniformly Sampled Data

Digital Frequency

Mag

nitu

de S

pect

rum

in d

BM=8

Y. C. Jenq 20

Estimation of Timing Errors - rm

A(k) =m=0,(M-1)[(1/M)exp(jrmoT)]e-jkm(2/M)

A(0)

A(1)A(2)

A(3)

Y. C. Jenq 21

Reconstruction of Digital Spectrum

-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5-60

-40

-20

0

20

40

60Spectrum Reconstruction from Non-uniformly Sampled Data

Digital Frequency

Mag

nitu

de S

pect

rum

in d

B

* : Reconstructed Spectrum, 10-bit quantization

Once the timing errors are known, can we reconstruct the correct digital spectrum?

Y. C. Jenq 22

Selecting Test Frequencies

A(0)

A(1)A(2)

A(3)

Higher frequency more sensitive to timing errorUsing FFT spurious harmonics should be on the binsWindowing function selection

Y. C. Jenq 23

Estimation of rm- Synchronous Case Residual Timing Error

timing offset error

RMS value beforeAdjust-ment

30%

20%

10%

5%

4x10-11

3x10-11

2x10-11

0.9x10-11

RMS value after

(4 bits)

RMS value after

(6 bits)

RMS value after

(8 bits)

RMS value after

(10 bits)

RMS value after

(∞ bits)

2.4x10-12

3.1x10-12

2.3x10-12

2.6x10-12

4.4x10-13

5.6x10-13

6.1x10-13

5.4x10-13

1.1x10-13

1.6x10-13

1.3x10-13

1.4x10-13

2.9x10-14

3.0x10-14

2.7x10-14

3.6x10-14

2.6x10-24

2.2x10-24

1.8x10-24

2.0x10-24

Residual timing errors are independent of initial timing errors!

Y. C. Jenq 24

Estimation of rm- Synchronous Case

4 6 8 10 12 14 1610

-6

10-5

10-4

10-3

10-2

10-1

Sensitivity of Timing Error Estimation Algorithm

Number of Effective Bits in A/D Converter

Res

idua

l Tim

ing

Err

or:

rm

* : Starting rm ~ 0.5

+ : Starting rm ~ 0.05

o : Starting rm ~ 0.005

Sensitivity to Quantization Noise in A/D Converter

Residual TimingError is relatively independent of initial timing error, but it is quite sensitive to the effective-bit of ADC

Y. C. Jenq 25

1

10-1

10-2

10-3

10-4

10-5

4 6 8 10 bits

Residual Timing Error: RMS rm

Residual Timing Error

One order of magnitude

improvement per 3 effective bits increase

Residual RMS rm ~ 10-3 at 7 Bits

Y. C. Jenq 26

Perfect Reconstruction of Digital Spectrum

Yih-Chyun Jenq, “Perfect Reconstruction of Digital Spectrum from Non-uniformly Sampled Signals,” IEEE Transactions on Instrumentation and Measurement, vol. 46, no. 3, 1997.

Y. C. Jenq 27

Reconstruction of Digital Spectrumwith Residual Timing Error

S/N ~ 20*log(1/) -16 dB

SNR = 6.02* (number of bits) + 1.76 dB

(Residual ~ (Initial /1000 at 7 Bits and improve one order of magnitude

per 4 bits increase

= standard deviation of rm

Reconstruction noise due to quantization error:

Reconstruction noise due to residual timing error:

Y. C. Jenq 28

Reconstruction of Digital Spectrumwith Residual Timing Error

Yih-Chyun Jenq, “Improveing Timing Offset Estimation by Aliasing Sampling,” IMTC’05, May 2005, Ottawa, Canada.