2001 Dynamic Testing High-Speed ADCs, Part Analog-to-digital

The Datasheet Archive - 100 Million Datasheets from 7500 Manufacturers.

2001 Dynamic Testing High-Speed ADCs, Part Analog-to-digital

Datasheet Thumbnail

Top Searches for this datasheet

CONVERSION/SAMPLING CIRCUITS BASESTATIONS WIRELESS INFRASTRUCTURE HIGH-SPEED SIGNAL PROCESSING
2001
Dynamic Testing High-Speed ADCs, Part
Analog-to-digital converters (ADCs) represent link between analog digital worlds receivers, test equipment other electronic devices. outlined Part this article series, number dynamic parameters provide accurate correlation dynamic performance expected from given ADC. Part this article series covers some setup configurations, equipment recommendations measurement procedures testing dynamic specifications high-speed ADCs. Additional Information: Defining Testing Dynamic Parameters High-Speed ADCs, Part following discussion setups procedures recommended testing high-speed data converters. includes software tools, hardware configurations, instruments data capture analysis needed test family 10-bit, +3V, high-speed data converters from Maxim. also warns traps encounter equipment selection, setup configuration, layout, FFT-based analysis performed with care. following topics covered:
Dynamic specifications definitions Board layout hardware configuration Power spectrum, bins, spectral leakage, window functions Software tools testing SNR, SINAD, THD, SFDR, TTIMD
Many approaches available acquiring output data from converters (not just highspeed ones) analyzing their dynamic performance. methods presented here represent proven approach, readers encouraged modify them necessary application hand.
Dynamic Specifications
those missed Part this discussion, following brief overview definitions mathematical descriptions important dynamic parameters high-speed ADCs.
Dynamic Parameter Signal-to-Noise Ratio (SNR) Signal-to-Noise Distortion Ratio (SINAD) Effective Number Bits (ENOB) Total Harmonic Distortion (THD) Spurious-Free Dynamic Range (SFDR) Two-Tone Intermodulation Distortion (TTIMD) Multi-Tone Intermodulation Distortion (MTIMD) Voltage Standing-Wave Ratio (VSWR)
Description/Definition SNRdB 6.02 1.763. SINADdB log10 (ASIGNAL[rms] ANOISE[rms]).
ENOB (SINAD 1.763) 6.02.
SFDR ratio expressed decibels amplitude fundamental (maximum signal component) value nextlargest spurious component, excluding offset. TTIMDdB log10 (AIMF_SUM[rms] AIMF_DIFF[rms]) AFUNDAMENTAL[rms]}. IMF_SUM IMF_DIFF TTIMD setup contain input tones only. MTIMDdB log10 (AIMF_SUM[ rms] AIMF_DIFF[rms]) AFUNDAMENTAL[rms]}. IMF_SUM IMF_DIFF MTIMD setup contain more than (usually four) input tones. VSWR where represents reflection coefficient.
Board Layout Hardware Requirements Test Setup
perform adequate dynamic tests high-speed data converters, should test board pre-assembled manufacturer follow data sheet's board-layout recommendations. This article considers layout requirements dynamic testing before delving into details hardware software. evaluation characterization board fast data converters (Figure must incorporate high-speed layout techniques (Figure 1c). usually replicate dynamic performance specified data sheet following these basic rules:
Locate bypass capacitors close device possible, preferably same side ADC, using surface-mount components achieve minimum trace length, inductance, capacitance. Bypass analog digital supplies, references, common-mode inputs with 0.1µF
ceramic capacitors parallel 2.2µF bipolar capacitor ground. Multilayer boards with separate ground power planes produce highest level signal integrity. Consider split ground plane arranged match physical location analog digital grounds ADC's package. impedance ground planes must kept possible, and, avoid possible damage latchup, their voltage differences both) must less than 0.3V. These grounds should joined single point, that noisy digital ground currents interfere with analog ground plane. ideal location this connection determined experimentally, point along between ground planes that produces optimum results. This connection achieved with low-value surface-mount resistor ferrite bead, direct short. alternative ground plane sufficiently isolated from noisy digital systems such downstream output buffer DSP), ground pins share same ground plane. Route high-speed digital signal traces away from sensitive analog traces. Keep signal lines short free turns. Always consider clock input analog input. Route away from actual analog inputs other digital signal lines.
Larger Image (PDF, 210k) Figure MAX1448 circuit schematic
Larger Image (PDF, 175k) Figure MAX1448 kit, optimized layout (component side)
Larger Image (PDF, 158k) Figure MAX1448 kit, optimized layout (solder side) proper test setup right test equipment1 necessary realize performance specified given converter (Figure 2b).
Figure System configuration test SNR, SINAD, THD, SFDR
Figure System configuration test two-tone following hardware2 proven extremely efficient therefore recommended test setup:
power supply: Hewlett Packard E3620A dual-supply 0-25V/0-1A. separate supplies analog digital nodes. Each must provide 100mA output drive current. Clock-signal function generator: Hewlett-Packard HP8662A. clock input device under test (DUT) accepts CMOS-compatible clock signals. This signal should have jitter fast rise fall times, because high-speed 10-stage pipeline, interstage conversion depends repeatability rising falling edges external clock. Sampling occurs falling edge clock signal, edge should have lowest possible jitter. Significant aperture jitter limits ADC's performance follows:
where represents analog input frequency time aperture jitter. Clock jitter especially critical undersampling applications. Input-signal function generator: Hewlett-Packard HP8662A. proper operation, this function generator should phase-locked clock-signal generator. Logic analyzer (LA): Hewlett-Packard HP16500C. Depending number points proposed FFT, able capture data using with less memory depth (such data record available HP1663C). Analog bandpass filter: Elliptical Function Bandpass Filter, series. Cutoff frequencies 7.5MHz, 20MHz, 40MHz, 50MHz. Digital multimeters (DMMs): Various Fluke, Keithley Instruments, multimeters (including hand-held HP2373A AC-powered HP34401A) were used setup check proper reference, supply, common-mode voltages.
Evaluating
simplify evaluation DUT, tested with performance-optimized, fully assembled tested, surface-mount board. Follow steps below configure setup operate this board. should complete connections before turning power supplies enabling function generators. Apply +3.0V analog power supply VAIN1 VAIN2, connect ground terminal AGND. Apply +3.0V digital power supply VDIN1 VDIN2, connect ground terminal DGND.
Verify that shunts installed jumpers (shutdown disabled) (digital outputs enabled). Connect clock function generator CLOCK connector. Connect output analog-signal function generator input bandpass filters. evaluate differential analog signals, verify that shunts installed pins jumpers JU4. Connect output bandpass filter DIFF connector. evaluate single-ended analog signals, verify that shunts installed pins jumpers JU4, connect output bandpass filter SINGLE connector. Connect logic-analyzer interface cables (pods) square header Turn both power supplies, verify +1.20V across test points with voltmeter. necessary, adjust potentiometer obtain +1.20V. Enable function generators. clock function generator maximum output amplitude (999mV suggested HP8662A) clock frequency fCLK 80MHz. analog signal function generator desired input tone, with amplitude between 10µV 999mV. Note that input amplitude frequency must selected according bandpass filter's corner frequency. Bandpass filters used evaluating high-speed data converters usually have very narrow passband. achieve optimum performance (depending filter type manufacturer, course), should input tone within corner frequency. Because filter attenuates generator's output signal, generator's amplitude slightly higher achieve desired full-scale input specification. proper operation, phase-lock (three, testing two-tone IMD) function generators. Synchronize with external clock signal from board, latch data clock's rising edge. Enable begin collecting data. Data stored floppy disk, LA's hard disk, data-acquisition (DAQ) board communicating through LA's HPIB bus. that necessary steps test setup hardware configuration have been completed system ready capture data from DUT, time select software tools data capture analysis:
LabWindows/CVIserves required data capture communications link between controller board. (The C-based program routine used this purpose will discussed this article.) MATLABis powerful tool that performs dynamic analysis captured data.
help understand MATLAB program routine analyzes graphs dynamic performance high-speed data converter, some power-spectrum basics
reviewed next section.
Power Spectrum, Bins, Spectral Leakage, Windowing
Fast Fourier Transform (FFT) power spectrum powerful tools measuring analyzing signals from captured data records. They capture time-domain signals, measure their frequency content, convert results convenient units, display them. perform FFT-based measurements, however, must understand issues calculations involved. Basic functions FFT-based signal analysis itself power spectrum. Both extremely useful measuring frequency content stationary transient signals. FFTs usually produce average signal's frequency content over time interval that signal acquired. Thus, FFTs always recommended stationarysignal analysis. Two-Sided Single-Sided Power-Spectrum Conversion Among most basic important computations signal analysis converting from two-sided single-sided power spectrum, adjusting frequency resolution, displaying spectrum. power spectrum usually returns matrix containing two-sided representation time-domain signal power frequency domain. values this matrix proportional amplitude squared each frequency component making time-domain signal. plot two-sided power spectrum usually contains both negative positive frequency components. Actual frequency-analysis tools, however, focus positive half frequency spectrum only, noting that spectrum real signal symmetrical around Negative frequency information therefore irrelevant. two-sided spectrum, half energy resides positive frequencies half negative frequencies. Therefore, convert from two-sided spectrum single-sided spectrum, discard second half matrix multiply every point (except two. Bins Frequency Resolution frequency range resolution x-axis spectrum plot (see program-code extraction below) depend sampling rate size data record (the number acquisition points). number frequency points lines power spectrum N/2, where number signal points captured time domain. first frequency line power spectrum always represents last frequency line found fSAMPLE/2 fSAMPLE/N. Frequency lines spaced even intervals fSAMPLE/N, commonly referred frequency (Figure
Figure representation frequency/FFT bins graph Bins also computed with reference ADC's sampling period: fSAMPLE/N 1/(N tSAMPLE)
example, with sampling frequency fSAMPLE 82.345MHz record length 8,192 data points, distance between each frequency line plot exactly 10.052kHz. (Refer Figure Defining Testing Dynamic Parameters High-Speed ADCs, Part calculations frequency axis (x-axis) proof that sampling frequency determines range bandwidth frequency spectrum. given sampling frequency, number points acquired time domain determines resolution frequency. increase resolution given frequency range, depth data record increased same sampling frequency (see following program-code extraction).
%Find signal number, %Span input frequency each side span=max(round(numpt/200),5); %Approximate search span harmonics each side spanh=2;
%Determine power spectrum %Find offset power Pdc=sum(spectP(1:span)); %Extract overall signal power %Vector/matrix store both frequency power signal harmonics Fh=[]; %The element vector/matrix represents signal, next element represents harmonic, etc. Ph=[];
Spectral Leakage Window Functions Window functions common analysis, their proper critical FFT-based measurements. following discussion spectral leakage stresses need select appropriate window function scale properly given application. accurately determine spectral leakage, however, enough adequate signal-acquisition techniques, convert two-sided power spectrum into single-sided one, rescale result. gain better understanding this term, should perform N-point spectrally pure sinusoidal input. Spectral leakage result assumption algorithm that time record precisely repeated throughout time that signals contained this time record periodic intervals corresponding length time record. However, nonintegral number cycles time record (fIN/fSAMPLE NWINDOW/ NRECORD) violates this condition causes spectral leakage (Figure (Refer Appendix Part Only cases guarantee acquisition integral number cycles:
Synchronous sampling with respect input tone capture transient signal that fits entirely into time record
most cases, however, application deals with unknown stationary3 input. This means there guarantee sampling integral number cycles. Spectral leakage distorts measurement spreading energy given frequency component over adjacent frequency lines bins. Selecting appropriate window function minimize effects this spectral leakage.
Figure effects windows spectral leakage fully understand given window function affects frequency spectrum, must take closer look frequency characteristics windows. Windowing input data equivalent convolving spectrum original signal with spectrum window. Even coherent sampling4, signal convolved with rectangular-shaped window uniform height. Such convolution shows typical sine-function characteristic spectrum. real-frequency characteristic window continuous spectrum consisting main lobe several side lobes. main lobe centered each frequency component signal time domain. Side lobes approach zero intervals each side main lobe. FFT, other hand, produces discrete frequency spectrum. continuous, periodic spectrum window sampled FFT, just would sample input signal time domain. What appears each frequency line value continuous, convolved spectrum each frequency line. frequency components original signal match frequency line exactly, case when acquire integral number cycles, only main lobe spectrum. Side lobes appear, because window spectrum approaches zero bin-frequency
intervals either side main lobe. time record does contain integral number cycles, continuous spectrum window shifted from main lobe center fraction frequency that corresponds difference between frequency component frequency lines. This shift causes side lobes appear spectrum. Thus, window's side-lobe characteristics directly affect extent which adjacent frequency components "leak into" neighboring frequency bins.
Window Characteristics
Before choosing appropriate window, necessary define parameters characteristics that enable users compare windows. Such characteristics include -3dB main-lobe width, -6dB main-lobe width, maximum side-lobe level, side-lobe rolloff rate (Table Side lobes window characterized maximum side-lobe level (defined maximum side-lobe level with respect main lobe's peak gain) side-lobe rolloff (defined asymptotic decay rate dB/decade dB/octave frequency) sidelobe peaks. Table Characteristics Frequently Used Window Functions
(Also refer MATLAB program code)
Window Type Window (Uniform) Hanning Hamming Flat
-3dB MainLobe Width
-6dB MainLobe Width
Maximum Side-Lobe Level -13dB -32dB -43dB -44dB
Side-Lobe Rolloff Rate 20dB/decade, 6dB/octave 60dB/decade, 18dB/octave 20dB/decade, 6dB/octave 20dB/decade, 6dB/octave
0.89 bins 1.21 bins 1.44 bins 2.00 bins 1.30 bins 1.81 bins 2.94 bins 3.56 bins
Selecting Right Window Different windows suit different applications. choose right spectral window, guess signal frequency content. signal contains strong interfering frequency components distant from frequency interest, should choose window whose side lobes have high-rolloff rate. strong interfering signals close frequency interest, window with maximum levels side lobe more suitable.
frequency band interest contains more signals close each other, spectral resolution becomes important. that case, window with narrow main lobe better. single frequency component which focus amplitude accuracy rather than precise location frequency bin, window with broad main lobe recommended. Finally, coherent sampling (instead window) recommended flat broadband frequency spectrum (see following program-code extraction).
window function used, input tone must chosen unique with %regard sampling frequency. achieve this prime numbers introduced %input tone determined fSAMPLE (Prime Number Data Record Size). relax this requirement, window functions such HANNING HAMING (see below) introduced, however fundamental resulting spectrum appears 'sharper' %without window functions. Doutw=Dout; %Doutw=Dout.*hanning(numpt); %Doutw=Dout.*hamming(numpt); %Performing Fast Fourier Transform Dout_spect=fft(Doutw); %Recalculate %Display results frequency domain with plot figure; maxdB=max(Dout_dB(1:numpt/2));
Hanning window function, which provides good frequency resolution reduced spectral leakage, yields satisfactory results most applications. Flat window good amplitude accuracy, wide main lobe provides poor frequency resolution more spectral leakage. Flat window lower maximum side-lobe level than does Hanning window, Hanning window faster rolloff rate. application consisting only transient signals should have spectral windows all, because they tend attenuate important information beginning sample block. case transient signal, should choose nonspectral window such Force Exponential window. Selecting appropriate window easy, signal content unknown start with Hanning characteristic. also excellent idea compare performance multiple window functions find most suitable given application. Table Signal Content Window Selection Advantages
Window Type Window (Uniform)
Signal Content
Window Characteristics
Broad-band random, Narrow main lobe, slow rolloff rate, closely spaced sine-wave poor frequency resolution signals Narrow-band random signals, nature content High maximum side-lobe level, unknown, sine-wave good frequency resolution, reduced combination sine-wave leakage, faster rolloff rate signals Closely spaced sine-wave signals Sine wave with need amplitude accuracy Good spectral resolution, narrow main lobe Good amplitude accuracy, wide main lobe, poor frequency resolution, more spectral leakage
Hanning
Hamming Flat
Dynamic-Range Specifications SNR, SINAD, THD, SFDR
With knowledge you've gained from preceding sections this article, following program-code extraction should easy understand. Based FFT, power spectrum, attention spectral leakage window functions, specifications SNR, SINAD, THD, SFDR calculated follows, using MATLAB: 10*log10(Ps/Pn) SINAD 10*log10(Ps/(Pn+Pd)) 10*log10(Pd/Ph(1)) SFDR 10*log10(Ph(1)/max(Ph(2:10))), where signal power, noise power, distortion power caused through 5thorder harmonics, Ph(1) fundamental harmonic power, Ph(2:10) harmonic power through 9th-order harmonics (see following program-code extraction power-spectrum level).
%Find harmonic frequencies power components spectrum har_num=1:10 %Input tones greater than fSAMPLE aliased back into spectrum tone>0.5 %Input tones greater than 0.5*fSAMPLE (after aliasing) reflected tone=1-tone; Fh=[Fh tone]; %For this procedure work, ensure folded back high order harmonics overlap %with signal lower order harmonics
Ph=[Ph %Determine total distortion power Pd=sum(Ph(2:5)); %Determine noise power format; AdB=20*log10(A) SINAD=10*log10(Ps/(Pn+Pd)) SNR=10*log10(Ps/Pn) disp('THD calculated from through order harmonics'); THD=10*log10(Pd/Ph(1)) disp('Signal Harmonic Power Components:'); HD=10*log10(Ph(1:10)/Ph(1))
Based MATLAB source code (see below), MAX1448 tested only data sheet specifications many other over- undersampling input frequencies well. achieved excellent dynamic performance under conditions.
%Example program routine generate plots determine dynamic performance high-speed dataconverter from data records taken with HP16500C Logic Analyzer %System. Data extracted through HPIB interface read into following MATLAB %program routine. same data extracted from controller interface %and simply copied floppy disk rather time-consuming way, possible. %Start MAX1448 Dynamic Performance Test Routine disp('HP16500C State Card'); filename=input('Type a:\filename Press RETURN HPIB Data Transfer: isempty(filename) filename 'listing'; fid=fopen(filename,'r'); numpt=input('Data Record Size (Number Points)? fclk=input('Sampling Frequency (MHz)? %MAX1448 10-bit data converter numbit=10;
%Discard first lines from data file, which contain data i=1:13, fgetl(fid); fclose(fid); v1=v1'; code=v1(:,2); %Display warning, when input generates code greater than full-scale (max(code)==2^numbit-1) (min(code)==0) disp('Warning: clipping!!!'); %Plot results time domain figure; plot([1:numpt],code); title('TIME DOMAIN') xlabel('SAMPLES'); ylabel('DIGITAL OUTPUT CODE'); %Recenter digital sine wave Dout=code-(2^numbit-1)/2; window function used, input tone must chosen unique with %regard sampling frequency. achieve this prime numbers introduced %input tone determined fSAMPLE (Prime Number Data Record Size). relax this requirement, window functions such HANNING HAMING (see below) introduced, however fundamental resulting spectrum appears 'sharper' %without window functions. Doutw=Dout; %Doutw=Dout.*hanning(numpt); %Doutw=Dout.*hamming(numpt); %Performing Fast Fourier Transform Dout_spect=fft(Doutw); %Recalculate %Display results frequency domain with plot figure; maxdB=max(Dout_dB(1:numpt/2));
%For TTIMD, following short routine, normalized -6.5dB full-scale. grid title('FFT PLOT'); xlabel('ANALOG INPUT FREQUENCY (MHz)'); ylabel('AMPLITUDE (dB)'); a1=axis; axis([a1(1) a1(2) -120 a1(4)]); %Calculate SNR, SINAD, SFDR values %Find signal number, %Span input frequency each side span=max(round(numpt/200),5); %Approximate search span harmonics each side spanh=2; %Determine power spectrum %Find offset power Pdc=sum(spectP(1:span)); %Extract overall signal power %Vector/matrix store both frequency power signal harmonics Fh=[]; %The element vector/matrix represents signal, next element represents %the harmonic, etc. Ph=[]; %Find harmonic frequencies power components spectrum har_num=1:10 %Input tones greater than fSAMPLE aliased back into spectrum tone>0.5 %Input tones greater than 0.5*fSAMPLE (after aliasing) reflected tone=1-tone; Fh=[Fh tone]; %For this procedure work, ensure folded back high order harmonics overlap %with signal lower order harmonics Ph=[Ph
%Determine total distortion power Pd=sum(Ph(2:5)); %Determine noise power format; AdB=20*log10(A) SINAD=10*log10(Ps/(Pn+Pd)) SNR=10*log10(Ps/Pn) disp('THD calculated from through order harmonics'); THD=10*log10(Pd/Ph(1)) disp('Signal Harmonic Power Components:'); HD=10*log10(Ph(1:10)/Ph(1)) %Distinguish harmonics locations within plot hold hold off;
Dynamic-Range Specifications, TTIMD
Two-tone tricky measurement, because additional equipment required power combiner combine input frequencies) contribute unwanted intermodulation products that falsify ADC's intermodulation distortion. must observe following conditions optimize performance, although they make selection proper input frequencies tedious task. First, input tones must fall into passband input filter. these tones close together (several tens hundreds kilohertz megahertz bandwidth), appropriate window function must chosen well. Placing them close together, however, allow power combiner falsify overall readings contributing unwanted 2nd- 3rdorder products (depending input tones' location within passband). Spacing input tones apart call different window type that less frequency resolution. setup also requires minimum three phase-locked signal generators. This requirement seldom poses problem test labs, generators have different capabilities matching frequency amplitude. Compensating such mismatches achieve (for example) -0.5dB two-tone envelope signal amplitudes -6.5dB will increase your effort test time (see following program-code extraction).
%For TTIMD, following short routine, normalized -6.5dB full-scale.
grid title('FFT PLOT'); xlabel('ANALOG INPUT FREQUENCY (MHz)'); ylabel('AMPLITUDE (dB)'); a1=axis; axis([a1(1) a1(2) -120 a1(4)]);
Conclusion
Besides points above, many other issues confront engineer trying determine dynamic range high-speed capturing signals analyzing them. Unfortunately, mistakes easily made spectral-measurement procedures. this task data acquisition analysis greatly eased understanding FFT-based measurement related computations, effect spectral leakage prevent necessary layout techniques equipment.
1The MAX1444/MAX1446/MAX1448 selected properly evaluate dynamic performance 10-bit, 80Msps MAX1448. 2For equipment suggested, substitute item more suitable your specific application. stationary signal present before, during, after data capture. 4Performing with apparent window function selected frequently referred performing with "uniform" "rectangular" window.
Literature Sources
MAX1448 data sheet, Rev. 10/00, Maxim Integrated Products. MAX1448 data sheet, Rev. 0/00, Maxim Integrated Products. Analog Integrated Circuit Design, Johns Martin, John Wiley Sons Inc., 1997. Low-Voltage/Low-Power Integrated Circuits Systems-Low-Voltage Mixed-Signal Circuits, SanchezSinencio Andreou, IEEE Press Marketing, 1999. Integrated Analog-to-Digital Digital-to-Analog Converters, Plasche, Kluwer Academic Publishers, 1994. Analog-Digital Conversion Handbook, Engineering Staff Analog Devices Inc., Prentice Hall Publishers, 1986. Mixed-Signal Design Techniques, Engineering Staff Analog Devices Inc., Printed Analog Devices, 2000. similar version this article appeared December 2000 issue Microwaves A220, January 2001
More Information MAX1180: QuickView Full (PDF) Data Sheet Free Samples MAX1181: QuickView Full (PDF) Data Sheet Free Samples MAX1182: QuickView Full (PDF) Data Sheet Free Samples MAX1183: QuickView Full (PDF) Data Sheet Free Samples MAX1184: QuickView Full (PDF) Data Sheet Free Samples MAX1185: QuickView Full (PDF) Data Sheet Free Samples MAX1186: QuickView Full (PDF) Data Sheet Free Samples MAX1190: QuickView Full (PDF) Data Sheet Free Samples MAX1195: QuickView Full (PDF) Data Sheet Free Samples MAX1196: QuickView Full (PDF) Data Sheet Free Samples MAX1198: QuickView Full (PDF) Data Sheet Free Samples MAX1420: QuickView Full (PDF) Data Sheet Free Samples MAX1421: QuickView Full (PDF) Data Sheet Free Samples MAX1422: QuickView Full (PDF) Data Sheet Free Samples MAX1444: QuickView Full (PDF) Data Sheet Free Samples MAX1446: QuickView Full (PDF) Data Sheet Free Samples MAX1448: QuickView Full (PDF) Data Sheet Free Samples MAX1449: QuickView Full (PDF) Data Sheet Free Samples

2001 Dynamic Testing High-Speed ADCs, Part Analog-to-digital

Top Searches for this datasheet

Other recent searches