Production Realization of MTPR Test on Low-Cost ATEfor OFDM Based Communication Devices

Ganesh Srinivasan & Abhijit Chatterjee &

Sasikumar Cherubal & Pramod Variyam

Received: 12 November 2008 /Accepted: 18 May 2010 /Published online: 5 August 2010# Springer Science+Business Media, LLC 2010

Abstract Multi-tone power ratio (MTPR) test is fastreplacing multiple single-carrier linearity and non-linearitytests for mixed-signal IC’s employed in broadband com-munication. Competitive cost models rule out the use ofexpensive automated test equipment that can performMTPR test in a specification compliant manner. In thispaper, deployment of a multi-tone dither based approach toperform MTPR tests on lower cost test platforms ispresented. The proposed method uses existing resourcesof a low-cost ATE to improve the linearity performance ofother resources required during the MTPR test. An ‘on-the-fly’ dither generation algorithm is developed to derive arobust dither signal accounting for variations typicallyencountered in production testing. Results obtained frommultiple test benches including ADSL mixed-signalCODEC ICs on TI’s internal low-cost platform is presentedto validate the proposed test method. Finally, statistical testdata obtained from conducted experiments is presented toevaluate the repeatability of the proposed approach.

Keywords Asymmetric digital subscriber line (ADSL) .

Multi-tone power ratio (MTPR) test .

Design-for-Test (DfT) . Low-cost automated test equipment(ATE) . Dither noise . High-volume production test .

Mixed-signal test

1 Introduction

The rapid surge in data transmission throughput enabled bycomplex multi-carrier modulation schemes has changed thetest matrix definition of mixed-signal ICs used for thesepurposes. Conventional specifications such as spurious freedynamic range, total harmonic distortion and two-tone inter-modulation distortion have become less meaningful forlinearity assessment of these mixed-signal ICs [7]. Also, thesespecifications carry little weight in establishing the perfor-mance of the ICs in broadband communication systems.Lack of IC test coverage more often than not impacts thetime-to-market cycle of the end-user system as additionaltests are required at other development stages. This has urgedsemiconductor manufactures to introduce modulated multi-tone tests such as MTPR to guarantee signal quality byemulating their real-time performance in the end-user system.

MTPR tests are typically used to test OFDM basedmixed-signal devices built for DSL, WLAN, WIMAXX,UWB and LTE applications. A single MTPR test canreplace multiple gain and linearity tests over a broadtransmission bandwidth. MTPR test involves complexalgorithms and modulation techniques that require high-quality data conversion devices (ADCs & DACs) andanalog front ends in the ATE [7]. Specialized test cardsintegrated to high-end ATE are used to perform these testsin a specification compliant manner. But specialized cardsdrive up the capital cost of the ATE and often are not

Responsible Editor: M. Margala

G. Srinivasan (*)Texas Instruments,Dallas, TX, USAe-mail: ganeshps@ti.com

G. Srinivasan :A. ChatterjeeGeorgia Institute of Technology,Atlanta, GA, USA

S. Cherubal : P. VariyamGeorgia Institute of Technology,Anora, LLC. 538 Haggard Street, STE 406 Plano,Texas 75074, USA

J Electron Test (2010) 26:405–417DOI 10.1007/s10836-010-5161-z

required for other class of devices during ATE re-use. Also,the capital costs of ATEs that can perform this test in aspecification compliant manner are often prohibitively highfor semiconductor manufactures to survive in competitiveand/or commoditized market spaces. ‘Low-cost ATEs’, acommon term used to symbolize ATEs with limitedperformance (in terms of digital pins, pin electronics,capture speeds, mixed signal & RF performance), oftenpossess limited MTPR measurement capability. This isbecause the ATE digitizer by itself lags behind thebroadband mixed-signal ICs in performance. The non-linearity of the digitizer introduces distortion components inmissing bins of the MTPR test thereby masking the deviceperformance. Hence, it is not possible to measure MTPR tothe required level of accuracy using a low-cost ATE.

In the past, little work has been done to lower the MTPRtest cost. In [9], a dithering technique to enhance theperformance of low-cost digitizers is presented. In this paper,the technique proposed in [9] is used to develop a robusthigh-volume production test solution for MTPR on low-costATE platforms. To this effect, a novel ‘on-the-fly’ multi-tonedither generation algorithm is developed. The generated‘optimal’ dither signal is added to the test signal to bedigitized using the DfT circuitry to randomize or smoothenthe non-linear static errors of the measurement equipmentthat limits the linearity performance. The multi-tone noise isa novel, high-volume production friendly form of dithernoise signal developed specifically with MTPR test, typicallow-cost ATE capability and production environment inmind. Also in this paper, the proposed approach is validatedon various test benches including TI’s proprietary low-costproduction test system to evaluate its robustness, the resultsof which are presented in detail. Also, results obtained fromstatistical studies are presented to demonstrate the good vs.bad screening capability of the proposed test solution.

2 MTPR Test Fundamentals

When performing the MTPR test, a multi-tone OFDMsignal, with specific missing tones, is supplied to thedevice-under-test (DUT) (Fig. 1) [7]. The location ofmissing tones in the signal spectrum is specified by theregulating standard for communication [1]. The amplitudesof all the existing tones are determined by the required

power spectral density. Due to the non-linearity of the DUT,distortion components raise the spectral energy in themissing frequency bins of the response spectrum. TheMTPR test specification value is determined by the ratio ofthe average power in existing tones to the maximum powerin any of the missing tones and is expressed in dB.

3 Effect of Dither on Periodic Signals

For periodic input signals such as the MTPR test waveform,static quantization errors (INL and DNL) are coherent dueto re-use of digitizer codes during quantization [4, 6]. Thesecoherent quantization errors add up to produce harmonicdistortion components in the spectrum of the quantizedsignal. Since the ATE digitizer is a non-linear device, itsdistortion components appear at the same integer multiplesof the test signal frequency (‘missing bins’ specifically),subsequently affecting the MTPR performance of the ATE.

In digital audio applications, it has been demonstrated thataddition of a noise signal (hereafter referred to as ‘dithersignal’) with certain temporal probability density function(PDF) along with the signal to be quantized (hereafter referredto as ‘input signal’) reduces or mitigates the harmonicdistortion components in single carrier applications [5, 6, 8].The reduction in harmonic distortion is due to the fact thatthe signal to be quantized is modified by the convolvingeffect of the dither noise to satisfy Widrow’s requirement[13]. When this happens, the errors due to quantization havea typical white noise characteristic and the probabilitydensity of this white noise is no longer correlated to theinput stimulus [13]. The modified PDF f x of the signal to bequantized can be expressed as, f x ¼ fx»fd , where fx is theactual PDF and fd is the PDF of the dither noise.

In [12], the impact of the use of dither signals withdifferent noise PDFs (Uniform, Gaussian) for reduction ofharmonic distortion components due to data converternonlinearity is presented. From [12], it can be concludedthat the performance of the dither signal has less depen-dence on the specific noise statistics and relies more on thechoice of optimum fd to meet Widrow’s requirements. Inthis work, concepts of dither and Widrow’s requirementsare used to develop a multi-tone dither noise with aGaussian PDF to reduce harmonic distortion in MTPRtests. The proposed dither noise modifies the transfercharacteristic of the ADC in the digitizer of the ATE forMTPR tests by modifying the PDF of the input signal. Thestatic errors in the transfer function of the ADC aresmoothened and this modified transfer characteristic QðxÞand corresponding static errors qeðxÞ can be expressed as:

QðxÞ ¼ QðxÞ»fdðxÞqeðxÞ ¼ qeðxÞ»fdðxÞ ð1Þ

ADSLDUT

DSL frequency band

Pow

er (

dBm

) Missing tones

Pow

er (

dBm

)

DSL frequency band

MTPRADSLDUT

DSL frequency band

Pow

er (

dBm

) Missing tones

Pow

er (

dBm

)

DSL frequency band

MTPR

Fig. 1 Pictorial description of the MTPR test procedure

406 J Electron Test (2010) 26:405–417

Where Q(x) is the observed ADC transfer characteristic andqe(x) are the observed ADC static errors.

The quantization error S of a data converter can beexpressed as,

SðxÞ ¼Xni¼1

u x� aið Þ � u x� bið Þ½ � ð2Þ

where n is the number of code edges, u is the unit stepfunction, and ai, bi are the codes for the actual and idealADC characteristics.

bi ¼ i� 1þ offset errorai ¼ bi þ DNLðiÞ ð3Þ

If the PDF of dither noise fd is even, then the modifiedquantization error at the output of the ADC can beexpressed as,

SðxÞ ¼ SðxÞ»fdðxÞ ¼Xni¼1

Zx�ai

x�bi

fdðuÞdu ð4Þ

For a Gaussian PDF this can be expressed as,

SðxÞ ¼ 0:5»Xni¼1

erfx� biffiffiffiffiffiffiffiffi2s2

p� �

� erfx� aiffiffiffiffiffiffiffiffi2s2

p� �� �

ð5Þ

where erf ðxÞ ¼ 2ffiffip

pRx0e�t2dt and σ is the standard deviation

of the noise PDF.As expressed in Eq. 5, increase in the standard deviation

of the noise signal decreases the quantization errors in theconverter. Since the PDF of the dither signal is Gaussian,increasing the standard deviation in turn increases the powerof the signal. This progressive improvement saturates at acertain power level above which the improvement is negli-gible for all practical purposes. It must also be noted thatlarger standard deviation values result in large values of peak-to-peak dither noise. Hence, exceeding a certain limit causesan increase in the noise floor, affecting the ADC performance.In certain cases, where expected improvement cannot beachieved due to constraints on maximum peak-to-peak dithernoise amplitudes, it becomes necessary to use subtractivedither [2–4]. It is also necessary that the dither signal ischosen in a manner that the sum of the dither signal and inputsignal does not exceed the dynamic range of the digitizer.

4 Proposed Approach for Low-Cost, High-VolumeAutomated MTPR Test

To improve MTPR test performance of the low-cost ATE,we propose to add a multi-tone out-of-band additive dithernoise signal with Gaussian PDF to the signal to bequantized. This dither signal N(t) can be expressed as:

NðtÞ ¼Xn

An sin nwnt þ fnð Þ ð6Þ

Where An is the amplitude of the nth tone, ωn is the angularfrequency of the nth tone, and fn is the initial phase of thenth tone. The fn of each tone is a random variable whosePDF is a Gaussian function to meet the noise requirementsof a dithering signal. It is essential to determine theoptimum amplitude of the dither signal for linearityimprovement, due to the tendency of large dither signalsto increase the noise floor of MTPR measurements. Theperformance of the dither signal in extending the ATElinearity is mainly dependent on the characteristics of itsPDF. The PDF of multi-tone signals is in turn determinedby its parameters: n, An, ωn, and fn.

The selection of the number of tones n, of the multi-tone signal is based on the power requirements of thedither signal. An increase in n causes a correspondingincrease in the power of the dither signal. The choice ofthe frequencies ωn of the tones is based on the availablesignal bandwidth. Also, the frequency of each tone in thedither signal is chosen to be an integer multiple of afundamental frequency to achieve coherent sampling. Thissampling frequency for the dither signal is calculated fromthe formula:

fnfsample

¼ Nwindow

Nrecordð7Þ

Where fin is the fundamental frequency of the multi-tonesignal, fsample is the ATE sampling frequency, Nwindow is aninteger number of cycles within the sampling window, andNrecord is the number of data points in the sampling window.

The proposed dither has a Gaussian PDF that is largelydependent on the amplitudes An of each individual tone(that jointly determine the power (variance) of the dithersignal due to its Gaussian nature) and less dependent on thevariance �s2

f of the random function used to generate thephases fn of individual tones. This can be explained fromthe following expression for power (variance) of the multi-tone dither signal N(t):

Var NðtÞð Þ ¼ E N2ðtÞ� � ð8Þ

N2ðtÞ ¼Xn;m

AnAm sin nwt þ fnð Þ sin mwt þ fmð Þ ð9Þ

Computing the left hand side of Eq. 9, the power(variance) of the dither signal can be expressed as:

Var N tð Þð Þ ¼ Pn

An2

2

� �1� e�s2

f

þe�s2

f 1� e�s2f

PnAn

2 cos 2nwtþ 2fnð Þ� �

ð10Þ

J Electron Test (2010) 26:405–417 407

where s2f is the variance of the Gaussian PDF used to

generate the random phase distribution of each dither tone.If the value of e�s2

is small, then it is clear from Eq. 10variance of phase distribution has negligible effect onpower (variance) of dither signal. Therefore, the power ofthe dither signal is dominated by the summation of thesquare of the amplitudes of the individual tones. Also,using a variance value of less than 3 degrees reduces thenoise characteristics of the multi-tone dither signals due toinsufficient variance of their phase values.

From Eq. 10, it can be concluded that the dither signalthat provided the required increase in ATE performancecan be generated by selecting the optimum amplitudes foreach of the individual tones of the dither signal. It isimportant to note that the choice of the dither signal isindependent of DUT performance and only improves ATEdigitizer performance for MTPR tests. However, the dithersignal needs to be generated in an ATE specific manner asdifferent ATEs suffer from different nonlinearities. Anapproach for generating optimal dither signals based onextracted nonlinearity errors (INL and DNL) of the ATEdigitizer is presented in [9]. This algorithm thougheffective for certain test cases is not the most suitableapproach for high-volume testing due to the followingreasons:

& Significant effort is spent in developing MATLABbased ADC models incorporating measured nonlinear-ity metrics for all ATEs employed. It is also difficult toincorporate tester-specific noise sources into this model.Hence, the generated MTPR tests may not be optimalacross all production test conditions.

& The method can suffer from low MTPR test repeat-ability as the statistics of the phase noise modeled inMATLAB may not be exactly the same as the ATEcase.

To overcome these limitations and to provide a robusttest solution for production deployment of MTPR tests onlow-cost ATE an ‘on-the-fly alternate dither generationalgorithm’ is proposed as follows.

4.1 ‘On-the-Fly’ Alternate Dither Generation Algorithm

As opposed to the prior approach, this algorithmcomputes the optimal dither noise on-the-fly in thehigh-volume test site during production test. Also, thederived dither noise is more accurate over variationstypically encountered in high-volume production. Thealgorithm is included in the main production test routineof ADSL devices and is executed ahead of high-volumetest procedures. This algorithm determines the optimalamplitude and phases of the multi-tones for improvedMTPR performance with a high-degree of repeatability.Since, the MTPR test stimulus for ADSL standardoccupies a bandwidth from 142 kHz to 1.1 MHz [1],the dither signals are placed in a bandwidth lower than100 KHz. As explained earlier, the selection of the numberof tones n, of the multi-tone signal is based on the powerrequirements of the dither signal. But emphasis needs tobe given to linearity improvement requirements andcapability of the AWG while choosing n. Using fewertones limits the randomness in the dither signal and usingmore than required tones can drive the AWG into a non-linear mode. In this application, a tone count of 100 waschosen as it provided the best trade-off between these twocases. Choosing other values near 100 only resulted in acorresponding change in An values.

4.1.1 On-the-Fly Alternate Dither Generation Algorithm

At first, the variables used in the subroutines and the mainprogram are defined below. In the following, ωn is a vectorcontaining a hundred frequency values ranging from100 Hz to 10 KHz in steps of 100 Hz, A0 is the initialamplitude value of the dither signal, Astep is the step inamplitude value for each iteration, MTPRB is the MTPRcomputed during bench characterization of ‘golden die’,MTPRi is the MTPR computed during ith iteration of ditheroptimization algorithm, and dither_noiseopt is the optimumdither noise (in terms of improving the ATE performancefor MTPR test) generated by the algorithm. Thesubroutines developed to support the main program aredescribed below.

RAND_GAUSS(m, v, n): This function creates aGaussian random variable vector with mean ‘m’,variance ‘v’ and size ‘n’.MTPR_test_routine: Subroutine to compute MTPRspecification of the DUT.Multi_tone_gen(n, A, ωn, fn): This function creates amulti-tone signal as given by Eq. 10 with ‘n’ tones ofequal amplitude ‘A’, frequency corresponding to nth

element of vector ‘ωn’ and phase corresponding to nth

element of vector fn.

Digital Source(Multi-tone TX input)

ADSLDUT

Dither noise generated using AWG

Low-cost digitizer(Signal capture

and FFT)

Fig. 2 Block diagram of proposed method to improve the digitizerperformance

408 J Electron Test (2010) 26:405–417

π

ω

ω

ω

The on-the-fly approach helps to optimize dither signalsin an ATE specific manner. The generated dither signalenables MTPR measurements with an improved degree ofrepeatability and accuracy on the low-cost ATE. Thecomputation of MTPR using the MTPR_test_routine isperformed on the average of ten captured test responses tofurther improve repeatability. Compared to the overall test

time required for each production lot, the additional timerequired to run this algorithm is negligible. Also, the goldendies are easily available at production sites as these are usedfrom time-to-time for calibration purposes.

The optimal dither signal obtained from the abovementioned algorithm is stored in the mainframe of theATE for MTPR specification measurement purposes.During MTPR test procedure in the main productionroutine, this dither noise is applied to the digitizer alongwith the ADSL device response in a cost-effective andreliable manner using the arbitrary waveform generator(AWG) resource already existent in the ATE and unused inthe conventional approach of this test. As the generatedmulti-tone dither signal is a noise-like signal, it is notrequired that the dither signal be synchronized to the MTPRinput stimulus. The dither signal could be a free-runningsignal and due to its out-of-signal bandwidth, multi-tonecharacteristics it can be reconstructed at the output of theADSL device accurately without any signal leakage.Figure 2 shows the block diagram of the proposed testmethod. The dither signal is added along with thetransmitted output of the ADSL DUT on the deviceinterface board (DiB). Suitable adder circuitry is designedand placed on the loadboard to enable this test. This addercircuitry poses negligible overhead in the real estaterequirements of the DiB.

The proposed ‘multi-tone dither signal’ has the follow-ing inherent advantages over previously proposed forms ofdither signals in [4, 6, 12], which facilitate their use inproduction test applications:

& The signal can be easily sourced using the arbitrarywaveform generator (AWG) of the ATE without anyrobustness issues.

& The signal has the characteristics of random noise butthat they can be accurately repeated.

-60 dB

Fig. 3 FFT plot of the low-cost ATE digitizer response to a TX MTPR test stimulus captured without dither noise (In all these FFT plots the X-axis represents the frequency in Hertz and the Y-axis represents the corresponding amplitude in dBm)

J Electron Test (2010) 26:405–417 409

& These signals are coherent and out-of-signal bandwidth.Hence, these signals will not cause any spectral leakagein the ADSL signal bandwidth during reconstruction.

& Large amplitude dither signals can be produced torandomize large non-linear errors.

5 Validation Results Obtained from Deploymentof Proposed Test Method on Low-Cost ATE

The proposed low-cost MTPR test method was validatedthrough multiple test benches. The MTPR test stimuli wereconstructed based on the ADSL standard. Specializedsoftware proprietary to TI was used for this purpose. Thefirst test bench was a 14-bit digitizer test card built usingoff-the-shelf components. The MTPR performance of thistest card with and without dither was evaluated for variousADSL MTPR waveforms and the results of these experi-ments are presented in this paper. In the second test bench,results obtained from deploying the proposed test approachon a state-of-the-art mixed-signal ADSL die on TI’s internallow-cost ATE platform is presented. The goal of the

deployment was to enable MTPR test of ADSL devices onthis low-cost platform in a specification compliant mannerthat was not possible earlier. In order to better understandthe test requirements of the ADSL device, a brief introduc-tion to ADSL fundamentals and testing is provided below.

5.1 ADSL Fundamentals

ADSL technology employs orthogonal frequency domainmultiplexing (OFDM) modulation in which its frequencyrange is divided into 256 sub-carriers with each sub-carrieroccupying 4.3125 KHz [10]. Each sub-carrier band acts asan independent channel and has its own stream of data. Therange of bins in the upstream and downstream channelsdepends on the mode of ADSL transmission (viz. Annex A,Annex A+, Annex J) [1]. Quadrature amplitude modulation(QAM) is used to modulate the individual carrier tones. De-pending on the dynamic range of each signal path, differentQAM modulation types can be used to transmit dataanywhere from 4 (2 bits) to 32768 (15 bits) in each carrier.

Depending on the initial phases of the DMT tones,ADSL signals with the same power can have different peak

-60 dB

Fig. 4 FFT plot of the low-cost ATE digitizer response to a TX MTPR test stimulus captured with dither noise

-62 dB

Fig. 5 FFT plot of the low-costATE digitizer response to a RXMTPR test stimulus capturedwithout dither noise

410 J Electron Test (2010) 26:405–417

voltage (Vpk) values in the time domain [11]. This is oftenthe case in ADSL transmission as it involves streamingreal-time random data. Hence, the ADSL standard requiresthe MTPR test of these devices to be performed fordifferent Vpk values. The parameter, PAR is used todifferentiate the discrete multi-tone waveforms based ontheir Vpk values and can be defined as,

PAR ¼ 20Log Vpk

, ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1

T

ZT0

f tð Þ2dt

vuuut0B@

1CA dBð Þ ð11Þ

where f(t) is the time domain discrete multi-tone waveformand T is the waveform period. To meet ADSL standardrequirements, MTPR tests are typically performed at 12 dB,14 dB and 16 dB PAR values for the TX channel. In thecase of RX channel, these tests are typically performed at10 dB and 16 dB [10].

5.2 Test Bench I: Validation Through Loopback of AWGand Digitizer Test Cards

In this case, the proposed approach was validated for twodifferent MTPR test stimuli cases, (1) TX upstream OFDM

stimulus with 16 dB PAR and (2) RX upstream stimuluswith 10 dB PAR. These signals were stored in the PC andsourced using a 16-bit AWG test card. This signal waslooped back and captured using a 14-bit digitizer test cardand stored in the PC. The FFT of the captured waveformwas used to compute and the MTPR specification in dB.The AWG and digitizer test cards were built using off-the-shelf components. Since, the goal of this case study was toevaluate the performance of dither and not to checkspecification compliance of MTPR; a reduced sample set(32 K samples) of the response signal was used to computethe FFT. Also the averaging function, which is normallyemployed in these tests, was not used in this case study. Theevaluation results obtained with and without dither noise fordifferent types of MTPR test stimuli are summarized below:

5.2.1 TX MTPR Test Stimulus with 16 dB PAR

MTPR test performed on the downstream channel of a COdevice (TX channel) is the most demanding among ADSLMTPR tests in terms of linearity requirements. This teststimulus contains 256 tones from 142 KHz to 1.1 MHz insteps of 4.3125 KHz. The amplitude values of all these tonesare set according to the requirements of the ADSL standard,except for a few tones that are set intentionally to zero(missing bins) for the MTPR test. The phases of these toneswere chosen such that a 16 dB PAR was achieved using thestimulus generation software. This signal was sourced using

Fig. 7 Dither signal for improving MTPR

Table 1 Summary of repeatability results for Repeated MTPR testsperformed on the same die

Mean Standard deviation

Without dither −63.796 0.541554

With dither −70.9893 1.158994

-74 dB

Fig. 6 FFT plot of the low-costATE digitizer response to a RXMTPR test stimulus capturedwith dither noise

J Electron Test (2010) 26:405–417 411

the AWG and captured by the digitizer using a loopbackoperation. 32 K samples of this signal were captured by thedigitizer and a 1024-point FFT was computed on this signal.

At first, the experiment was conducted without the dithersignal to evaluate the performance of the digitizer card. Theobserved FFT plot in this case is shown in Fig. 3. Toimprove the linearity performance of the digitizer, suitabledither noise (generated using the proposed algorithm) wasadded along with the MTPR signal and sourced using theAWG. The corresponding FFT plot for the response signalcaptured by the digitizer is shown in Fig. 4. Comparing theperformance of the ATE with and without the dither signalit can be observed that an improvement of ∼7 dB can beachieved using the proposed method. This improvementwas due to the randomization of the nonlinear quantizationerrors in the ATE digitizer. The corresponding spurioustones caused by these nonlinear errors in the missing binsof the MTPR test were lowered to the noise floor. It canalso be observed from Fig. 4 that the MTPR measurements

are now limited by the noise floor and not by the linearityof the digitizer.

5.2.2 RX MTPR Test Stimulus with 10 dB PAR

The loopback approach was also evaluated for the MTPRtest stimulus applied to the upstream channel of a CO device(RX). These signals typically exhibit a higher dynamic rangecompared to the downstream MTPR tests, as they havefewer carrier tones. These tests are less demanding in termsof linearity of the digitizer, but nevertheless difficult to per-form on low-cost ATE platforms. As in the case of the TXMTPR test, the performance of the digitizer was evaluatedby the loopback operation with and without the dither signal.Similar settings as in the previous experiment were used forthe signal capture and FFT computation purposes.

The FFT plot of the test response captured using thedigitizer without the dither signal is shown in Fig. 5. Againthe dither generation algorithm was used to generateoptimal dither noise. The FFT plot of the signal capturedupon addition of this dither signal to the MTPR test signalis shown in Fig. 6. In this experiment, it was found that theMTPR performance of the digitizer for the RX test could beimproved from −62 dB MTPR to −74 dB using theproposed approach. As in the previous experiment, furtherimprovements were limited by the noise floor and not bythe linearity of the digitizer.

5.3 Test Bench II: Validation on ADSL CO CODEC Device

In this case study, the proposed test method was used toperform specification compliant MTPR test of ADSL COCODEC devices on TI’s internal low-cost ATE platform.The device interface board (DiB) a.k.a. loadboard was

Fig. 8 Repeatability results of MTPR measurements performed onsame die (1) without and (2) with dither noise

-63 dB

Fig. 9 TX MTPR for ADSL CODEC DUT measured on a low-cost ATE

412 J Electron Test (2010) 26:405–417

designed to accommodate a differential active summercircuitry to add the dither signal to the device test responsesignal, along with the other standard design-for-test (DfT)circuitry. TI’s THS 4503 was used as the summer amplifier,since its harmonic distortion components was less than−100 dBc at 1.1 MHz (largest frequency component in theADSL TX input spectrum).

In this experiment, a downstream MTPR input stimulus(TX channel) with 16 dB PAR was applied to the DUT. Thedigital source of the ATE was used to source this signal tothe TX channel of the CODEC device. The ATE digitizerwas used to capture the analog response of this device at asampling frequency of 16 MHz. 65 K samples of the timedomain signal at the transmitter output port of the CODECdevice was captured and the FFT was computed on theaverage of ten such captured signals. This FFT plot ispresented in Fig. 9. The MTPR computed from this FFTplot was limited to −63 dB due to the non-linearities in thedigitizer. These devices typically measure about −75 dBMTPR on bench. Also, the ADSL standard requires thesedevices to pass the MTPR specification test of at least −65 dBto be employed for ADSL data transmission purposes.

To improve the performance of the digitizer and henceenable specification compliant MTPR tests, optimum dithersignal was generated. This dither waveform is presented inFig. 7. 2048 samples of the dither signal were generatedand sourced using the AWG of the ATE. The higher thanexpected peak-to-peak dither noise (typically with 1 LSB)is due to the fact that multi-tone dither noise occupieslimited spectrum in the frequency domain, as opposed topreviously proposed white noise type dither signals. Hence,more power is concentrated in fewer tones. This dithernoise signal need not be synchronized to the MTPRwaveform as this is a noise signal and its efficiency isoptimized for the non-linearities in the digitizer and isindependent of the input waveform. This signal was addedto the output of the DUT on the DiB and captured using thedigitizer of the ATE. Figure 10 shows the correspondingFFT plot of the captured signal along with the dither signal.Comparing the FFT plots in Figs. 9 and 10, an improve-ment of ∼7 dB in MTPR performance can be observed.Also from Fig. 10, it can be observed that distortioncomponents have been reduced to the noise floor and nofurther reduction is possible as we have already reached thenoise floor of the ATE.

5.3.1 Repeatability Analysis of MTPR Tests PerformedUsing Proposed Test Method on the Same Die

Repeatability analysis of the proposed test method wasperformed on same device as part of the optimization

-71 dB

Fig. 10 FFT plot of the low-cost ATE digitizer response to a TX MTPR test stimulus captured with dither noise

Table 2 Summary of repeatability results for MTPR tests performedover different dies

Die ID Without dither With dither

1 −63.89 −70.342 −63.52 −70.263 −63.67 −71.904 −64.19 −70.965 −64.01 −70.626 −63.62 −69.657 −63.83 −70.388 −63.35 −71.80

Table 3 Summary of repeatability results for ‘failed’ device

Mean Standard deviation

Without dither −62.6823 0.549424

With dither −62.2979 0.621273

J Electron Test (2010) 26:405–417 413

algorithm to select the optimal dither signal and also topresent the validation results in a statistical manner. Table 1summarizes the repeatability results of MTPR measure-ments made on the device with and without the dither noisesignal. From the table, an improvement of more than 7 dBcan be observed in the mean value of measured MTPR,using the proposed test method. Figure 8 presents thehistogram plot of these measurements. In Fig. 8, the MTPRvalues on the left hand side refer to those measured with theoptimal dither noise and the values on the right hand siderefer to those measured without the dither noise. Thestandard deviation (repeatability) of the MTPR valuemeasured without the dither noise was ∼0.5 dB, whichwas achieved through averaging of ten signal captures. Thestandard deviation value for the MTPR measurement withdither was ∼1.1 dB. This variation is due the fact that wehave reduced all the distortion components in the digitizerto the noise floor. Hence the variations are due to therandom nature of the noise floor which adds additionalvigor to the proposed method (Figs. 9 and 10).

5.3.2 Repeatability Analysis of MTPR Tests PerformedUsing Proposed Test Method over Different Dies

The proposed test method for MTPR test was repeated foranother set of eight devices to evaluate its robustness acrossdifferent devices that exhibit different MTPR values due toinherent process variations. The MTPR specificationcomputed for these devices using high-end bench equip-ment was ∼−75 dB. Table 2 presents the MPTR measure-ments made on these devices with and without the optimaldither noise. In all these MTPR measurements, the ATEdigitizer was used to capture the DUT response at asampling frequency of 16 MHz. 65 K samples of the timedomain signal at the transmitter output port of the CODECdevice was captured and the FFT was computed on theaverage of ten such captured signals. From the table, it canbe observed that an improvement of ∼7 dB in MTPRperformance can be achieved for each of these devices.

Under normal circumstances, i.e. without the additionof dither noise, all these devices would have failed the

Fig. 14 FFT computed on the digitized response signal as observed atthe output of the 14-bit digitizer model after the addition of generateddither noise signal

Fig. 13 Time domain samples vs. amplitude of the dither noise signalin volts

Fig. 12 FFT computed on the digitized response signal as observed atthe output of the 14-bit digitizer model

Fig. 11 FFT computed on the response of the DUT model to the1 MHZ input stimulus

414 J Electron Test (2010) 26:405–417

MTPR specification test in the high-volume productionroutine as the ADSL standard requires a minimum of−65 dB MTPR for the downstream tests; although thesedevices exhibit −75 dB MTPR performances when testedwith high-end bench equipment. Hence, it can beconcluded from these experiments that addition of suitablemulti-tone dither helps to perform more accurate MTPRtests on ADSL devices in a specification compliantmanner over a high-volume of devices using low-costATE.

5.3.3 Verification of Proposed Test Method on ‘Failed’ IC

To verify that the proposed multi-tone dither noise onlyreduces the distortion added by the digitizer and not thatof the device, the proposed approach was implementedon a device that had failed the MTPR specificationduring bench measurements. Table 3 summarizes therepeatability results obtained for this device with andwithout the addition of dither noise. The mean andstandard deviation values with the dither noise show verygood resemblance to the respective values without dither.Hence, it can be concluded that the dither only reduces thedistortion added by the digitizer (i.e. only forwardrandomization).

6 ATE Digitizer Improvement for Other Linearityand Modulation Applications

The proposed on-the-fly dither noise generation algorithmand application concept can also be applied to improvethe ATE digitizer performance for other narrow and wideband tests like SDFR, THD, IIP3, and EVM. Since, the

goal of the developed approach is to mitigate the effect ofquantization error in the digitizer during these tests; it isalso independent of the modulation scheme employed. Todemonstrate this, we apply the approach to improve theperformance of a 14-bit digitizer during single carrierSFDR tests.

The DUT employed in this experiment was a 14-bitCODEC device with better performance than the 14-bitADC in the digitizer model. The models for both theCODEC and digitizer was developed in MATLAB withsuitable gain, offset, INL and DNL errors in their codeedges. The INL error at the nth code edge was calculated asthe cumulative sum of DNL errors up to the nth code andcan be represented as:

INL ið Þ ¼Xi

j¼1

DNL ið Þ ð12Þ

These static nonlinearities were incorporated in themodels by altering the quantization code edges accord-ingly with suitable gain and offset errors. The pseudocode used to model the digitizer operation is presentedbelow:

A 1 MHz sine wave with 4096 sample points at asampling rate of 40.96 MHz was used at the input stimulusto the CODEC device. The FFT calculated on the timedomain response of the CODEC to the 1 MHZ inputstimulus is shown in Fig. 11. This plot highlights the power

Table 4 Summary of power levels observed in the fundamental andhighest harmonic components for the three cases

Fundamental(dB)

Highest HarmonicComponent (dB)

CODEC Response (Ideal case) −21.58 −87.90Digitizer capture without dither −21.58 −80.55Digitizer capture with dither −21.58 −85.10

Table 5 Summary of SFDR results observed in statistical study

Mean (dB) Standard deviation (dB)

CODEC SFDR (Ideal case) 65.9374 0.2609

Measured SDFR without dither 59.2595 0.5690

Measured SDFR with dither 64.2355 0.9774

Handler &Handler kit

Operational Cost

Infrastructure

ATESystem

30%30%

10% 20%

Ha

30%30%

10% 20%

Fig. 15 Typical test cell investment breakdown

J Electron Test (2010) 26:405–417 415

in dB observed at the carrier frequency and thecorresponding first, second and third harmonic components.

The FFT computed on the CODEC response as capturedby the digitizer model is shown in Fig. 12. As observedfrom the plot, the power of the harmonic distortioncomponents added by the digitizer model degrades theSFDR performance. Since, the digitizer model has worseINL and DNL errors they mask the actual deviceperformance. The proposed on-the-fly dither generationalgorithm was used to generate the optimum dither noise.The generated noise signal is presented in Fig. 13. Thefrequency span of this signal is from 10 KHz to 500 KHzwhich is out-of-band to the test stimulus.

The generated dither signal was added to the signal to bedigitized (response of the CODEC device) and the resultantsignal captured by the digitizer model is shown in Fig. 14.As observed from the figure, the SDFR performance hasbeen improved as the power of the largest distortioncomponent in the spectrum has been lowered. A summaryof the power levels in the fundamental tone and highestharmonic components observed in each of these three casesis presented in Table 4.

To perform a statistical study, 200 different instances ofthe digitizer were generated by tweaking the quantizationerrors (INL and DNL). The experiments were repeated onall these instances and the SFDR was calculated for eachof the measurement cases. Table 5 summarizes the meanand standard deviation values for the SFDR specificationvalues measured with and without dither noise. Animprovement of ∼5 dB was observed in the mean valuesof the SDFR measurements with less than 1 dB standarddeviation value.

7 Impact on Test Cost

The proposed method results in a ‘5X’ reduction in testcost of ADSL devices, while maintaining the same testtime. The primary cost reduction comes from the capitalcost difference of the low-cost vs. high-end ATEemployed in traditional approaches. Also, the low-costATE can be re-used for other products that do not requirethe specialized test cards and do not impact their test cost.Moreover, using low-cost test cards to perform thesemeasurements enables us to add more multi-site capabilityto the ATE. As shown in the pie chart of Fig. 15, the ATEcapital cost typically occupies only 30% of the overall testcell investment. Hence, utilization of all other high-volume test resources (specifically test handlers) throughhigher multi-site efficiency results in an 80% lower testcell investment (based on octal site).

8 Conclusion

This paper presents novel test method to enable productiondeployment of MTPR tests on low-cost ATE. In the proposedapproach, optimal multi-tone dither noise generated using theproposed ‘on-the-fly’ generation algorithm is added to the testsignal to improve the linearity performance of digitizers thatlimit the range of possible MTPR measurements. Validationexperiments performed on multiple test benches includingstate-of-the-art mixed-signal ADSL ICs on TI’s internal low-cost test platform show promising and robust results whichmore than significant impact on test cost. Applications inother areas are presented to highlight the broad scope of theproposed approach

Acknowledgment The authors wish to thank the ‘TTC-MAKE’team at Texas Instruments, Dallas for their support in evaluating theproposed methodology on the low-cost ATE platform. The authorsalso wish to thank M. Teklu, C.P. Wong, and D. Guidry for theirvaluable input to complete this work.

References

1. ADSL2 and ADSL2plus-The new ADSL standards (2003)[Online]. Available: http://www.dslforum.org

2. Babu BNS, Wollman HB (1998) Testing an ADC linearized withpseudorandom dither. IEEE Trans Instrum Meas 47(4):839–848

3. Badzmirowski K, Jackiewicz B (1999) Effect-iveness ofvarious dithering methods for linearization of high-resolutionover sampling A/D converters. In Proc. 3rd InternationalConference on Advanced A/D and D/A Conversion Techniquesand their Applications, (Conf. Publ. No. 466), July 1999, pp175–178

4. Brannon B Overcoming converter non-linearities with dither,analog devices application note AN-410

5. Carbone P, Petri D (1994) Effect of additive dither on theresolution of ideal quantizers. IEEE Trans Instrum Meas 43(3):389–396

6. Carbone P, Narduzzi C, Pedri D (1994) Dither signal effects onthe resolution of nonlinear quantizers. IEEE Trans Instrum Meas43(2):139–145

7. Dibattista L, Oberhauser C (2003) Meeting test challenges forADSL. [Online]. Available: http://www.eetasia.com

8. Lipshitz SP, Wannamaker RA (1992) Quantization and dither: atheoretical survey. J Audio Eng Soc 40(5):355–375

9. Srinivasan G, Cherubal S, Variyam P, Teklu M, Wang CP, GuidryD, Chatterjee A (2005) Accurate measurement of multi-tonepower ratio (MTPR) of ADSL devices using low-cost testers. InProc. European Test Symposium 2005, May 2005, pp 68–73

10. Starr T, Cioffi JM, Silverman PJ (1998) Understanding digitalsubscriber line technology. Prentice Hall, New Jersey, ch. 2–3

11. Van der Ouderaa E, Schoukens J, Renneboog J (1988) Peak factorminimization of input and output signals of linear systems. IEEETrans Instrum Meas 37(2):207–212

12. Wagdy MF (1996) Effect of additive dither on the resolution ofADC’s with single-bit or multi-bit errors. IEEE Trans InstrumMeas 45(2):610–615

416 J Electron Test (2010) 26:405–417

13. Widrow B, Kollàr I, Liu M (1996) Statistical theory ofquantization. IEEE Trans Instrum Meas 45(2):353–361

Ganesh Srinivasan is a product engineer in the High-Volume Analog(HVAL) Group at Texas Instruments. Srinivasan has an MS and a PhDin electrical and computer engineering from the Georgia Institute ofTechnology. Email: ganeshps@ti.com

Abhijit Chatterjee is a professor at the Georgia Institute ofTechnology, where he currently directs a mixed-signal design and testprogram. Chatterjee has an MS in electrical engineering and computerscience from the University of Illinois at Chicago, and a PhD inelectrical and computer engineering from the University of Illinois atUrbana-Champaign. Email: chat@ece.gatech.edu

Sasikumar Cherubal is the director of technology at Anora LLC. Heholds Doctoral and Masters degrees from Georgia Institute ofTechnology School of Electrical and Computer Engineering, and aBachelors degree in Electronics and Electrical CommunicationsEngineering from Indian Institute of Technology, Chennai. He has

been working on testing of mixed signal and RF devices for the past10 years. He has authored fifteen refereed publications and holds 5U.S patents.

Pramod Variyam is a founder and managing partner at Anora, LLC.He holds Doctoral and Masters degrees from Georgia Institute ofTechnology School of Electrical and Computer Engineering, and aBachelors degree in Electrical Engineering from Indian Institute ofTechnology, Chennai.

Prior to founding Anora, Dr. Variyam was the director of productdevelopment engineering at WiQuest Communications. At WiQuest,he productized two generations of ultra-wide-band chipsets. Hearchitected and implemented low-cost test solutions for 8 GHzsingle-chip OFDM based CMOS UWB radio with 2.5 GHz PCIeand USB2.0 host interfaces.

Previously, he worked at Texas Instruments where he wasinvolved in productization of cutting edge semiconductor chips intohigh volume manufacturing. As manager of product developmentengineering in TI’s Broadband Communications Group, Dr. Variyamwas responsible for productizing three generations of DSL centraloffice chips. As a part of TI’s Wireless Terminal Business Unit, hereleased integrated power management and voice band CODECdevices into high volume manufacturing.

He has authored more than twenty refereed publications, holdsfive U.S. Patents and has three more pending.

J Electron Test (2010) 26:405–417 417