SIGNAL MODEL BASED APPROACH TO JOINT JITTER AND NOISE DECOMPOSITION
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
SIGNAL MODEL BASED APPROACH TO JOINT JITTER AND NOISE DECOMPOSITION White paper | Version 01.00 | Adrian Ispas, Julian Leyh, Andreas Maier, Bernhard Nitsch First published at DesignCon 2020
ABSTRACT We propose a joint jitter and noise analysis framework for serial PAM transmission based on a parametric signal model. Our approach has several benefits over state-of-the-art methods. First, we provide additional measurements. Second, we require shorter sig- nal lengths for the same accuracy. Finally, our method does not rely on specific symbol sequences. In this paper, we show example measurement results as well as comparisons with state-of-the-art methods. AUTHORS BIOGRAPHY Adrian Ispas received Dipl.-Ing. and Dr.-Ing. degrees in Electrical Engineering and Information Technology from RWTH Aachen University, Aachen, Germany, in 2007 and 2013, respectively. Since 2013, he has been with Rohde & Schwarz GmbH & Co. KG, Munich, Germany, where he is now a Senior Development Engineer in the Center of Competence for Signal Processing. Julian Leyh obtained his M.Sc. degree in the field of Electrical Engineering from TU Munich in 2017 specializing in communications engineering. He then began his pro- fessional career as a member of the Center of Competence for Signal Processing at Rohde & Schwarz. His current focus is on time domain high-speed digital signal analysis. Andreas Maier received an M.S. degree in Electrical Engineering and Ph.D. degree from the Karlsruhe Institute of Technology (KIT), Germany in 2005 and 2011, respec- tively. In 2011, he joined Rohde & Schwarz, Munich, Germany. His areas of expertise are digital signal processing and signal integrity. He is a lecturer at the Baden-Wuerttemberg Cooperative State University Stuttgart, Germany. Bernhard Nitsch is the director of the Center of Competence of Signal Processing at Rohde & Schwarz in Munich, Germany. He and his team focus on signal processing algo- rithms and systems in the area of signal analysis, spectral analysis, power measurement, security scanners and oscilloscopes. Rohde & Schwarz products contain software, FPGA or ASIC based signal processing components developed in the Center of Competence. He joined Rohde & Schwarz in 2000 and holds a PhD degree in Electrical Engineering and Information Technology from Darmstadt University of Technology, Germany and an M.S. degree in Electrical Engineering from Aachen University of Technology, Germany. He is author of one book and seven technical papers and holds more than 25 patents. ACKNOWLEDGEMENT We want to thank Thomas Kuhwald for initiating the development of the joint jitter and noise framework, Guido Schulze for motivating us to submit the paper to DesignCon and finally Andrew Schaefer, Guido Schulze and Josef Wolf for their valuable feedback on the paper. 2
CONTENTS 1 Introduction....................................................................................................................................................4 2 Signal model...................................................................................................................................................5 3 Jitter and noise decomposition.....................................................................................................................6 3.1 Source and analysis domains..........................................................................................................................6 3.2 Decomposition tree..........................................................................................................................................7 4 Joint jitter and noise analysis framework....................................................................................................8 4.1 Estimation of model parameters......................................................................................................................9 4.2 Analysis domains...........................................................................................................................................10 5 Measurements..............................................................................................................................................11 5.1 Statistical values and histograms..................................................................................................................11 5.2 Duty cycle distortion and level distortion......................................................................................................11 5.3 Autocorrelation functions and power spectral densities..............................................................................11 5.4 Symbol error rate calculation........................................................................................................................11 5.5 Jitter and noise characterization at a target SER..........................................................................................12 6 Framework results........................................................................................................................................13 6.1 Example analysis...........................................................................................................................................13 6.2 Signal length influence.................................................................................................................................19 7 Comparison with competing algorithms.....................................................................................................19 8 Conclusion....................................................................................................................................................22 9 References....................................................................................................................................................23 Rohde & Schwarz White paper | Signal model based a ecomposition 3 pproach to joint jitter and noise d
1 INTRODUCTION The identification of jitter and noise sources is critical when debugging failure sources in the transmission of high-speed serial signals. With ever increasing data rates accompa- nied by decreasing jitter budgets and noise margins, managing jitter and noise sources is of utmost relevance. Methods for decomposing jitter have matured considerably over the past 20 years; however, they are mostly based on time interval error (TIE) measurements alone [1, 2]. This TIE-centric view discards a significant portion of the information present in the input signal and thus limits the decomposition accuracy. The field of jitter separation was conceived in 1999 by M. Li et al. with the introduction of the Dual-Dirac method [3]. The Dual-Dirac method was augmented and improved over the next two decades. Originally, it was meant to isolate deterministic from random jit- ter components based on the probability density of the input signal’s TIEs. It uses the fact that deterministic and random jitter are statistically independent and that determin- istic jitter is bounded in amplitude, while random jitter is generally unbounded. Three years later, M. Li et al. reported that their Dual-Dirac method systematically overestimates the deterministic component [4]. Despite this flaw, modelling jitter using the Dual-Dirac model has maintained significance in commercial jitter measurement solutions due to its simplicity [5]. Throughout the years, additional techniques were added to the original Dual-Dirac method to separate additional jitter sources such as intersymbol interference (ISI), peri- odic jitter (PJ) and other bounded uncorrelated (OBU) jitter. For example, PJ components can be extracted using the autocorrelation function [6] or the power spectral density [7, 8] of the TIEs, while the ISI part of deterministic jitter can be determined by averag- ing periodically repeating or otherwise equal signal segments [13]. Yet another method for estimating ISI makes use of the property that ISI can be approximately described as a superposition of the effect of individual symbol transitions on their respective TIE [12]. Once the probability density function of one jitter component is known, any second com- ponent can be estimated from a mix of the two by means of deconvolution approaches, as long as the components are statistically independent of each other [9, 10, 11]. Collectively, over 40 IEEE publications and more than 50 patents can be found on the topic of jitter analysis alone. Despite this, applications in industrial jitter measurements commonly use a combination of the previously described methods [14, 15, 16], all based solely on TIEs. In this paper, we first introduce a parametric signal model for serial pulse-amplitude mod- ulated (PAM) transmission that includes jitter and noise contributions. The key to this model is a set of step responses, which characterizes the deterministic behavior of the transmission system, similar to the impulse response in traditional communication sys- tems. Based on the signal model, we propose a joint jitter and noise analysis framework that takes into account all information present in the input signal. This framework relies on a joint estimation of model parameters, from which we readily obtain the commonly known jitter and noise components. Therefore, we provide a single mathematical base yielding the well-known jitter/noise analysis results for PAM signals and thus a consistent impairment analysis for high-speed serial transmission systems. 4
known jitter/noise analysis results for PAM signals and thus a consistent impairment analysis for high-speed Additionally, we provideserial deeptransmission systems. system insight through the introduction of new mea- surements, such as what-if signal reconstructions based on a subset of the underlying Additionally, we provide deep system insight through the introduction of new impairments. These reconstructions enable the visualization of eye diagrams for a selec- measurements, such as what-if signal reconstructions based on a subset of the underlying tion of jitter/noise impairments. Thesecomponents, thereby reconstructions enableallowing informed the visualization ofdecisions about eye diagrams for the relevance a selection ofofthe selected components. Similarly, we determine selective symbol error jitter/noise components, thereby allowing informed decisions about the relevance of rate (SER) the and bit error rate (BER) extrapolations to allow fast calculation of (selective) selected components. Similarly, we determine selective symbol error rate (SER) and bit peak-to-peak jitter errorandratenoise (BER) amplitudes at low extrapolations error rates. allowing for a fast calculation of (selective) peak-to-peak Signal Model-Based Approach to a Joint Jitter & jitter and noise amplitudes at low error rates. The proposed framework is inherently able to perform accurate measurements even relativelyNoise Decomposition The proposed framework is inherently able to perform accurate measurements even for using relatively short input signals. This is due to the significant increase in information short input signals. This is due to the significant increase in information extracted extracted from the signal. Furthermore, our approach has no requirements regarding spe- from the signal. Furthermore, our approach has no requirements regarding specific input cific inputsequences, symbol symbol sequences, such as compliance such as predefined predefined patterns. compliance On patterns. On random the contrary, the contrary, or the randominput scrambled or scrambled inputencountered data typically data typically encountered in real-world in real-world scenarios is ideallyscenarios suited to is theide- ally suited to the framework. framework. Abstract 2 SIGNAL MODEL 2 Signal Model The proposed joint jitter I. Theproposed andI NTRODUCTION andnoise noiseanalysis analysis framework framework is based is based on aon a signal signal modelmodel for for serial serial PAMPAM data data II.This transmission. transmission. S IGNAL This model M modelODEL assumes assumes the total the total signal, signal, i.e. received i.e., the the received signal signal containing all containing allcomponents, components, totobebe ∞ Signal Model-Based ∆ [k] · h (tApproach toy a+ Joint Jitter &(1) y(t) = − T [k] − [k], s) + (t). (1) s sr clk h −∞ v Noise the vector Decomposition k=−∞ (1) Here,s denotes Here, of the transmitted PAM symbol sequence denotes the vector of the transmitted PAM symbol sequence and and Signal Model-Based h Δ(t Ts[k] Δ[k] − [k] − sr s [k], s[k = s)Approach 1] 1 clk the = hsymbol – denotes the step response for the symbol s h T [k] −atto (t −difference – the vector t T [k]at symbol symbol a [k], {s[k], time response for the symbol vector at time , clk sr , Joint index difference the (t, clk s) Jitter k∆h.[k]}). Moreover, at reference ℎ , symbol h clock time & (2) index . ssr denotes the reference clock time at symbol clk sr the step k Noise y Decomposition at symbol index , and the initial signal value. The disturbance sources are split into index , and –∞ the initial signal value. The disturbance sources are split into horizontal horizontal (time) and vertical (signal level) parts: is the total horizontal source (time) and vertical (signal disturbance and is ϵ [k] NPlevel) (h) −1 parts: h is the total horizontal source disturbance and Abstract the total vertical source disturbance. Contrary to traditional ϵ v(t h [k] = h,P [k] + h,OBU ) is the total vertical [k] + h,R systems, source [k] = we useA disturbance. stepsin(2πf Contrary to Th,l [k]of+ traditional anφimpulse communication sys- (3) communication a h,l responseh,linstead h,l ) + response h,OBU [k]to+model h,R [k] tems, we use a step the total signal. This is due response to instead the fact of thatan animpulse impulse response response to is model neither the able total to signal. ensure l=0 This signalis due to the under continuity fact that an−1 non-linear impulse (in theresponse signal levelis neither domain) able to ensure horizontal signal continuity disturbances nor under able tononlinear (insymbol N the signal I. P (v) I NTRODUCTION level domain) horizontal disturbances nor able to repro- reproduce Abstract transition Abstract asymmetries. The step response includes transmitter, v (t) = v,P (t) + v,OBU duce channel,(t) +and symbol v,R (t) = receiver transition possibly asymmetries. A II. S IGNAL v,l sin(2πf effects, TheM allstep ODEL t + φresult response ofv,lwhich v,l )includes +inv,OBU (t) + v,Rchannel transmitter, a data-dependent (t). signal and (4) possibly receiver effects, all of which result in a data-dependent signal disturbance,the disturbance, i.e., inter-symboll=0 interference. Moreover, the step response depends on i.e. symbol vector in order toMoreover, account for effects like symbol transition dependencies in the s in intersymbol III.∞system. J ITTER interference. InI. N OISE D ECOMPOSITION II NTRODUCTION AND I.the NTRODUCTION the step response depends on the symbol vector transmission order to account for effects following, we assume the step response to depend only on the like symbol transition dependencies in the transmission sys- A. Source and Analysislast Domains y(t) = and the symbol ∆slastII. [k] II. · hS − the M (t transition: SsrIGNAL symbol Tclk [k] ODEL Mstep −response h [k], s)to+depend y−∞ + (t). (1) tem. In the following, we assume IGNAL ODEL onlyv on the last symbol B. Decomposition Treeand the lastk=−∞ symbol transition: ∞ ∞ y(t) h sr = (t − T = 1 clk ∞ ∆ [k] ∞ [k]h··[k], ∆ss− [k] h s)− hsrsr (t n=T (t −−Tclk sr[k] Thclk h [k], −+ [k] s) y−∞{s[k], h [k], + v (t). ∆s [k]}). (2) (1) (1) (2) y(t) (t∂− clk [k] − h [k], s) + y−∞ + v (t). yR(h) (t) = k=−∞ k=−∞ ∆s [k] · n hsr (t − Tclk [k], {s[k], ∆s [k]}) · (−h,R [k])n . (5) n! ∂t Information Classification: General n=1 k=−∞ We thus have up to N ⋅ (N PAM NP (h) PAM−1 – 1) step responses for a signal of PAM order N PAM . h sr (t − T clk [k] − h [k], s) = h sr (t − T [k] − h [k],ϵ {s[k], [k] ∆ϵsv[k]}). (t ) fur- (2) h The (t − total T horizontal [k] h [k] = h,P [k] sr+ h,OBUclk[k] + − h,R ∞ and [k], h [k] =s)vertical = h sr source (t − T disturbances clk Ah,lclk [k] − sin(2πfh,l [k], {s[k], h and h Th,l [k] + φh,l ∆ s ) + h,OBU [k] +are , [k]}). respectively, h,R [k] (2) (3) ther decomposed as ∂ ỹR(h) (t) = − ∆s [k] · l=0hsr (t − Tclk [k], {s[k], ∆s [k]}) · h,R [k]. (6) NP∂t(v) −1 k=−∞ NP (h) NP −1 (h) −1 h [k] [k] = v (t) = = h,P [k] (t) [k] + v,P + + h,OBU [k] [k] + v,OBU (t) + + h,R [k] [k] = v,R (t) = = A Av,l sin(2πf Ah,l sin(2πf sin(2πfh,l T v,l t + φ Th,l [k] [k] + +φ v,l )h,l + v,OBU φ )) + (t) + +v,Rh,R + h,OBU [k] (t). [k] + [k] [k] (3) (4) (3) h h,P h,OBU h,R h,l h,l h,l h,l h,OBU h,R l=0 l=0 l=0 III. N OISE D ECOMPOSITION J ITTER NP (v)AND NP −1 (v) −1 A. (t) = Source (t) and + (t) Analysis (t) + Domains (t) = A sin(2πfv,l t + φv,l ) + v,OBU (t) + v,R (t). (4) vv (t) = v,P v,P + v,OBU v,OBU (t) + v,R v,R (t) = Av,l v,l sin(2πfv,l t + φv,l ) + v,OBU (t) + v,R (t). (3) (4) l=0 B. Decomposition Tree l=0 III. J ITTER AND N III. J ITTER AND OISE D N OISE D ECOMPOSITION ECOMPOSITION A. A. Source Source and and Analysis Analysis Domains Domains ∞ ∞ 1 & Rohde ∂ n | Signal model based a pproach to joint jitter and noise d ecomposition Schwarz White paper 5 B. B. Decomposition y Decomposition Tree Tree R(h) (t) = ∆ s [k] · n hsr (t − Tclk [k], {s[k], ∆s [k]}) · (−h,R [k])n . (5) n! n=1 k=−∞ ∂t
The horizontal and vertical periodic components ϵh,P [k] and ϵv,P (t ), respectively, are characterized by the amplitudes Ax,l , the frequencies fx,l and the phases ϕx,l for l = 0, … , NP(x) – 1 and x = {h,v}, respectively, and Th,l [k] denotes the relevant point in time for the horizontal periodic component l at symbol k. The terms ϵh,OBU [k] and ϵv,OBU (t ) des- ignate other bounded uncorrelated (OBU) 1) components, and the random components ϵh,R [k] and ϵv,R (t ) denote additive noise for the horizontal and vertical case, respectively. At this point, we do not impose any statistical properties on the random components. 3 JITTER AND NOISE DECOMPOSITION As introduced in section 2, various effects cause disturbances in transmitted data. The transmitter and the channel have the most influence on the step responses that deter- mine the data-dependent disturbance of the signal (intersymbol interference). Periodic and OBU disturbances make up the remaining deterministic and bounded components. Finally, there are random and unbounded disturbances such as thermal noise. Apart from being deterministic and bounded, very little information is available about OBU com- ponents at the receiver end. Therefore, we omit OBU components from now on. Their influence will be visible in the extracted random components. 3.1 Source and analysis domains With the exception of data-dependent components, all disturbances are either of hori- zontal or vertical origin. The horizontal components constituting ϵh [k] originate at the transmitter, whereas the vertical components in ϵv (t ) may also be added in the channel or at the receiver end. The data-dependent components have their origin in the combination of the data, i.e. the transmitted symbols, and the step responses that overlap to build the signal. We thus define the source domain to be either “horizontal”, “vertical” or “data”. Analyzing the disturbances in the time domain is referred to as jitter analysis. Timing errors with respect to a reference clock signal are determined and analyzed. This is usu- ally done by means of the TIE. However, the disturbances can also be analyzed in the signal level domain, which is referred to as noise analysis. In this case, signal level errors with respect to reference levels are determined and analyzed at symbol sampling times given by a reference clock signal. Fig. 1 depicts the definition of the TIE and the level error (LE). The choice between jitter analysis and noise analysis is a choice of the analysis domain, which we accordingly define to be either “jitter” or “noise”. Fig. 1: TIE and LE definition Signal level LE Reference level Threshold level TIE Sampling time Clock time Transition point 1) The word “other” in OBU refers to the nonperiodicity of the disturbance and the word “uncorrelated” to the nonexistence of any correlation between the disturbance and the symbol sequence. 6
hsr (t − Tclk [k] − h [k], s) =NP h(h) −1− Tclk [k] − h [k], {s[k], ∆s [k]}). sr (t (2 h [k] = h,P 3.2 [k] +Decomposition h,OBU [k] +tree h,R [k] = Ah,l sin(2πfh,l Th,l [k] + φh,l ) + h,OBU [k] + h,R [k] The decomposition tree of total jitter l=0 and total noise is depicted in Fig. 2. It is important to NP (h) −1 remember that, as shown in Fig. NP2, (v)horizontal −1 disturbances also have an influence in the h [k] = h,P [k] + noise [k] domain h,OBU + and [k] = vertical h,R disturbancesA h,l sin(2πf in the jitter Th,lt[k] h,ldomain. + φh,lno Thus ) +matter h,OBU [k] + ah,R whether jit-[k] (3 v (t) = v,P (t) + v,OBU (t) + v,R (t) = Av,l sin(2πf v,l + φv,l ) + v,OBU (t) + v,R (t). ter or a noise analysis is performed, l=0 l=0 all disturbance components need to be accounted for, i.e. for any analysis domain, all NP (v) −1 source AND N OISE domains are relevant. Moreover, it is generally notv,OBU possible III. to directly J v,R (t) mapITTER = properties Av,linsin(2πf D ECOMPOSITION a source (4 v (t) = v,P (t) + (t) + v,ldomain t + φv,lto )+measurements v,OBU (t) + in an(t). v,R analy- A. Source and Analysis sis domain.Domains Consider, e.g. thel=0 case of the horizontal random source disturbance ϵh,R [k] only. Its effect on the signal level and thus the noise domain can be described by a Taylor B. Decomposition Tree III. J ITTER AND N OISE D ECOMPOSITION series of ϵh,R [k] around the mean. For the zero-mean case, we obtain A. Source and Analysis Domains ∞ ∞ B. Decomposition Tree 1 ∂n yR(h) (t) = ∆s [k] · n hsr (t − Tclk [k], {s[k], ∆s [k]}) · (−h,R [k])n . (4) n=1 n! k=−∞ ∂t ∞ 1 ∞ ∂n yR(h) (t) = ∆ ∞s [k] · hsr (t − Tclk [k], {s[k], ∆s [k]}) · (−h,R [k])n . n ∂ (5 n! The first-order approximation ∂t of equation ỹR(h) n=1(t) = − k=−∞ ∆s [k] · hsr (t(4)−isTgiven by{s[k], ∆s [k]}) · h,R [k]. clk [k], k=−∞ ∂t ∞ ∂ ỹR(h) (t) = − ∆s [k] · hsr (t − Tclk [k], {s[k], ∆s [k]}) · h,R [k]. (5) (6 k=−∞ ∂t We recognize that the horizontal random source disturbances are transformed to the noise domain in a nonlinear manner that depends on the step responses and their deriva- tives 2). Moreover, the inverse step of determining the source disturbance based on the noise disturbance is nontrivial. Fig. 2: Jitter and noise decomposition tree a) Jitter analysis Jitter analysis Unbounded Random horizontal jitter RJ(h) Random jitter RJ Random vertical jitter RJ(v) Total jitter TJ Bounded Periodic horizontal jitter PJ(h) Periodic jitter PJ Periodic vertical Deterministic jitter DJ jitter PJ(v) Data-dependent jitter DDJ Noise analysis Unbounded Random horizontal noise RN(h) Random noise RN Random vertical noise RN(v) Total noise TN Bounded 2) For jointly stationary ϵh,R [k] and s [k], and Tclk [k] = kTs with the constant symbol period Ts, the first-order Periodicapproximation horizontal (5) is cyclostationary. Moreover, for PAM signals of order NPAM > 2 and Gaussian distributed ϵh,R [k], ỹR(h) (t ) isPN(h) noise not Gaussian dis- tributed since the symbol difference Δ s [k] can take on values with different amplitudes. Periodic noise PN Periodic vertical Rohde & Schwarz White paper | Signal Deterministic model based a noise DN noise PN(v) pproach to joint jitter and noise ecomposition 7 d Data-dependent
Deterministic jitter DJ jitter PJ(v) Data-dependent Fig. 2: Jitter and noise decomposition tree jitter DDJ b) Noise analysis Noise analysis Unbounded Random horizontal noise RN(h) Random noise RN Random vertical noise RN(v) Total noise TN Bounded Periodic horizontal noise PN(h) Periodic noise PN Periodic vertical Deterministic noise DN noise PN(v) Data-dependent noise DDN 4 JOINT JITTER AND NOISE ANALYSIS FRAMEWORK We propose a framework that encompasses a joint jitter and noise analysis and is based on the signal model introduced in section 2. Contrary to state-of-the-art methods, which transition from a signal to a TIE representation as a first step, our approach is based directly on the total signal. It thus does not discard any information present in the (total) signal by using a condensed representation like TIEs do. The core of our framework operates independently of the analysis domain (jitter or noise) and considers all source domains (horizontal, vertical and data). The jitter and noise results are derived from a joint set of estimated model parameters. Conceptually, the first step of the framework is to recover a clock that makes it possible to decode, i.e. recover, the data symbols. Based on the recovered clock signal and the decoded symbols, a joint estimation of model parameters is performed. Specifically, a coarse, partly spectrum based estimator recovers rough estimates of the frequencies and phases of the vertical periodic components as well as their number. These are then used for a fine, least-squares estimator that jointly determines the step responses and the peri- odic vertical components. The horizontal periodic components are similarly estimated. These estimates, along with the recovered symbols and clock, allow synthesis of signals with only selected disturbance components, e.g. the data-dependent signal. We then use the total and synthesized signals to transition to the analysis domains and calculate jitter and noise values, i.e. TIE and LE values, respectively. Fig. 3 shows a block diagram of the framework. 8
Fig. 3: Joint jitter and noise analysis framework Clock and data recovery Estimation: step responses Signal synthesis Jitter analysis and periodic components Coarse estimation Fine estimation SIGDD SIGT EST_PV_ SYNTH_DD CLK_REC DECODE EST_SR_PV TIEX COARSE TIE SIGP(v) SYNTH_PV TIEP(h) + R SIGDD + P(h) EST_PH_ EST_PH SYNTH_DD COARSE LEX LE Noise analysis 4.1 Estimation of model parameters The fine estimator that performs a joint estimation of the model parameters is at the core IV. J OINT of the framework and isJ derived N OISE as follows. ITTER AND A NALYSIS First, we need toFtreat the remaining horizon- RAMEWORK A. Estimation tal disturbances ϵ of Model Parametersh, rem[k] as additive noise for the estimator. Their influence in the signal level domain is quantified as in (5) by linearizing (1) with respect to ϵh, rem[k]: ∞ IV. IV. J OINT J ITTER AND N OISE A NALYSISJ OINT J ITTER F RAMEWORK AND N (t) ≈ ỹh,rem (t) = − OISE A∆ ˇNALYSIS F RAMEWORK ˇ yh,rem s [k] · h(t − Ťclk [k], {š[k], ∆s [k]}) · h,rem [k] (6) A. Estimation of Model Parameters Model Parameters k=−∞ ∞ where š [k]Nis the decoded P (v)−1 symbol, Δ̌ s [k] the corresponding symbol difference, IV. ˇ s [k] · h(t − Ťclk t,J{š[k], h((v) šOINT [k]= ,∆ˇ Jss [k]}) [k] ∞ ITTER ) the AND [k]N OISE impulse ˇ A NALYSIS response, and Τ̌ clk F (7) [k]RAMEWORK an estimated clock time. rem (t) ≈ ỹh,rem (t) = − y ∆ v,P (t) ≈ ỹP[k], (t) ≈ ỹh,rem (t) = −(t) · A v,lˇsin(2π fv,l t + φ̌v,l ) + Av,l (∆ h,rem ˇ f,v,l 2πt + ∆s [k] · h(t − Ťclk [k], {š[k], ∆s [k]}) · h,rem [k] ∆φ,v,l ) cos(2π fˇv,l t + φ̌v,l ) (7) A. Estimation k=−∞ ofh,rem Model Parameters l=0k=−∞ NP (v)−1 IV. J OINT J ITTERBy ANDfurther approximating N OISE A NALYSIS theFvertical RAMEWORK sinusoidal components, we obtain ∞ Estimation (t) = ofAv,l Model sin(2πParameters fˇv,l t + φ̌v,l ) + Av,l (∆ NPf,v,l (v)−1 2πt +∆ ∞ ) cos(2π fˇv,l t + φ̌v,l ) (8) ˇ y(t) ≈ ỹ(t) = ∆ˇ φ,v,l [k] · h (tˇ−s [k]Ť · [k], {š[k], l=0 v,P (t) ≈ ỹPy(v) = ≈ ỹh,remA (t)(t) h,rem − fˇv,l t∆ (t)v,l=sin(2π s sr + φ̌clk h(t ) + − A Ťclk∆ v,l (∆ s [k]}) [k], +∆ {š[k], ˇy−∞ + ỹ· P (v) (t)[k] [k]}) f,v,l 2πt + ∆sφ,v,l ) cos(2π + v,R (t) + ỹh,rem h,remfˇv,l t + φ̌v,l ) (7) (t). (8) k=−∞ v,l (7) k=−∞ ∞ l=0 ∞ IV. J OINT J ITTER AND N OISE A NALYSIS F RAMEWORK ˇ s [k] · y ∆ hh,rem Ťclk≈ sr (t −(t) ˇ s= {š[k],(t) [k],ỹh,rem ∆ [k]}) ˇ sỹ[k] − + y−∞ ∆ + P (v)·(t) h(t+− v,RŤ(t) clk [k], {š[k], + ỹh,rem ˇs(9) (t). ∆ [k]})· h,rem [k] (7) k=−∞ A. Estimation ∞ oftheModel NP (v)−1 with frequenciesParameters fv,l = ƒ̌v,l + Δf,v,l andĥthe sr phases ϕv,l = ϕ̌ v,l + Δϕ,v,l split into coarse and k=−∞ y(t) ≈ ỹ(t) ≈ỹP= ˇ [k] · hAsr (tparts − Ťclk flˇv,l t0,+…φ̌, v,l = {š[k], N∆ˇ)x̂s+= 1.v,l ŷThis =2πt +A + ỹP+y ˇv,lỹth,rem (9) (8) v,P (t) (t) = ∆sremaining (v) v,l sin(2π for[k], [k]}) P(v) – A + (∆yf,v,l −∞ yields −∞ the ∆(t) following (v) φ,v,l+) cos(2π v,R (t) f + approximation φ̌(t). + of the v,l ) total NP (v)−1 sr k=−∞+ signal: ĥ l=0 ∞ p̂ x̂ = ŷ−∞ = A y ≈A − (10) ˇ [k] · h(t − ˇ s [k]}) · h,rem [k] v,P (t) ≈ ỹP (v) (t) = p̂Av,l sin(2π ∞ fˇv,l tyh,rem + φ̌v,l(t)) + ỹh,rem v,l (∆ (t) f,v,l =2πt ∞ + ∆ ∆ φ,v,ls) cos(2π fv,l t clk ˇ Ť +[k], φ̌v,l{š[k], ) ∆ (8) ĥsr ˇ k=−∞ y(t) ≈l=0 ỹ(t) = ∆ˇ s [k] · hsr (t −ŷŤDD [k], clk(t) = {š[k], ∆∆ ˇ s [k] + s [k]}) ·+ĥsry(t−∞ −+Tcdr ỹP[k], (v) (t) ∆ˇ s+ v,R [k]) +(t) + ỹh,rem (t). (10) ŷ−∞ (9) (8) ∞ x̂ = ŷ −∞ = A y ŷDD (t) = ∞ ˇ s [k] · ĥsr (tk=−∞ ∆ − Tcdr [k], ∆ ˇ s [k]) + ŷ−∞NP (v)−1 k=−∞ p̂ (11) NP (v) −1 y(t) ≈ ỹ(t) = k=−∞∆ ˇ s [k] · hsr (tv,P− (t) Ťclk≈ [k], ỹP{š[k], (v) (t) ∆ˇ= s [k]}) + y A−∞ v,l + sin(2πỹ fˇ (t)t + φ̌ P (v)v,l v,R v,l )(t) ++ Av,lỹh,rem (∆f,v,l (t).2πt + (9)∆φ,v,l ) cos(2π fˇv,l t + φ̌v NP (v) −1 The coarse estimates ŷ (t) and = the ĥ above  approximations sin(2π f ˆ t + can φ̂ be). used to perform a (fine) joint k=−∞ ∞ P (v) l=0 sr v,l v,l v,l ˆ least-squares estimation of · −∞the step − responses and the vertical periodic components: ŷP (v) (t) = Âv,l sin(2π fv,l t + φ̂v,l ). (t) = ŷDD x̂∆ˇ= s [k] ŷ ĥsr (t l=0 = TA + (12)[k], ˇ cdr y ∆s [k]) + ŷ−∞ (10) (11) l=0 ∞ p̂ ĥsr k=−∞ ˇ ˇ s [k]}) + y−∞ + ỹP (v) (t) + v,R (t) + ỹh,rem ( y(t) ≈ ỹ(t) =ŷ−∞ P N ∆ s A y sr − [k] −1 +· h (t T [k],−{š[k], Ťclk[k] T ∆ (10) x̂ = = (v) [k] (9) (v) Tsig,T [k] − Tsig,DD+P (v) [k] sig,T (13) sig,DD+P k=−∞ ∞ ˆ ŷP (v) (t)p̂=  v,l sin(2π fv,l t + φ̂v,l ). ˇ s [k] · ĥsr (t − Tcdr [k], ∆ ˇ s [k]) + ŷ−∞ (12) ins B. Analysis Domains ŷDD (t) = l=0 ∆ (11) ∞ with the estimate vector x̂ containing the Nĥ k=−∞ sr srestimates of the step responses in the ˇ vector ĥ , the (v) −1 signalˇvalue estimate NPinitial x̂ = ŷ −∞ , and =A + the y estimates (11) of the frequencies, TIEX [k] ŷDD = T(t)sig,X=[k] − Tcdr∆ [k]s [k] ·srĥsr (t −Tsig,T Tcdr [k][k],− ∆T s [k]) + (14) ŷ−∞ [k] (13) TIE [k] sig,DD+Pˆ = T [k] − T [k] + φ̂v,lp̂).components in the vector p̂ . (v) (12) k=−∞phases, ŷP (v) (t)and = amplitudes Âv,lofsin(2π the verticalfv,l t periodic X sig,X cdr B. Analysis DomainsNP (v) −1 l=0 TIET [k] TIED = (t) ŷP (v) =[k] + TIER [k] sin(2π fˆ t + φ̂ ). ∞ (15) (12) d ecomposition 9 v,l Rohde & Schwarzv,l v,l paper | Signal ˇ(16) ˇ sjitter TIEDD+P (h) [k] = TIEDD [k]l=0 + TIEP (h) [k] ŷ White (t) = TIE Tsig,T [k] − Tsig,DD+P DD T [k] ∆ TIE =s model [k] ·D [k] ĥbased [k] sr (t +a − TIE pproach TcdrR to[k] [k], joint ∆ [k])and+noise ŷ−∞ (13) (v) TIEDD+P (v) [k] = TIEDD [k] + TIEP (v) [k] TIE [k] XTIEDD+P = T k=−∞ sig,X(h) [k] [k] − (17) = TTIE cdr [k] DD [k] + TIE P (h) [k] (14)
y(t) ≈ y(t) ≈ ỹ(t) ỹ(t) = = ∆ss[k] ∆ [k] ·· hhsrsr(t (t − − TTclk [k], {š[k], clk[k], {š[k], ∆ ∆ss[k]}) [k]}) + + yy−∞ −∞ + + ỹỹPP(v) (v)(t) (t) ++ v,R v,R(t) (t) + + ỹỹh,rem (t). (9) h,rem(t). (9) ∞ IV. J OINT J ITTER AND N OISE A NALYSIS F RAMEWORK k=−∞ k=−∞ ≈ ỹ(t) =of Model )stimation ∆ˇ s [k] · hsr (t − Ťclk [k], {š[k], ∆ ˇ s [k]}) + y−∞ + ỹP (v) (t) + v,R (t) + ỹh,rem (t). (9) Parameters IV. J OINT TheJmatrix ITTERAAND + is the A NALYSIS Npseudoinverse OISE of the F RAMEWORK observation matrix and the vector y contains k=−∞ ĥ ĥsrsr A. Estimation ofIV. Model J OINTParameters J ITTERsamples ANDofNthe OISE total A x̂ x̂ signal. NALYSIS == ŷŷ−∞Based on these F RAMEWORK A++yy estimates and the recovered (or input) clock (10) (10) −∞ = =A signal ∞ Τ cdr [k] , the data-dependent (DD) and the periodic vertical (P(v)) signals can be imation of Model Parameters synthesized ˇ ĥ as sr p̂ p̂ ˇ yh,rem (t) ≈ ỹh,rem (t) = − ∆ s [k]· h(t −+Ťclk [k], {š[k], ∆s [k]}) · h,rem [k] (7) x̂ = ŷ ∞ −∞ = A y (10) yh,rem (t) ≈ ỹh,rem (t)∞= − p̂ k=−∞ ∞ ∞∆ ˇ s [k] · h(t − Ťclk [k], {š[k], ∆ ˇ s [k]}) · h,rem [k] (7) ŷŷDD ˇˇ [k] ·· ĥĥsrsr(t(t − − TTcdr ˇ ˇ (11) (11) DD(t) (t)ˇ k=−∞ == ∆∆ [k] ˇ [k], [k], ∆ ∆ss[k]) [k]) + + ŷŷ−∞ yh,rem (t) ≈ N (t) = −∞ ỹh,rem P (v)−1 ∆s [k] · h(t −ssŤclk [k], {š[k], cdr ∆ s [k]}) · h,rem [k] −∞ (7) k=−∞ k=−∞ v,P (t) ≈ ỹP (v) (t) = ˇ t + φ̌v,l ) + Av,l (∆f,v,l 2πt + ∆φ,v,l ) cos(2π fˇv,l t + φ̌v,l ) (8) NPA sin(2π ˇfsv,l k=−∞ ŷDD (t) =v,l (v)−1 ∆ [k] · N ĥNPsrP (v)−1 (t (v) −1 − Tcdr [k], ∆ ˇ s [k]) + ŷ−∞ (11) = fˇv,l t + φ̌ ). ∆φ,v,l ) cos(2π fˇv,l t + φ̌v,l ) (8) (12) l=0 v,P (t) ≈ ỹP (v) NP(t)(v)−1= A ŷŷPPv,l(v) k=−∞ (v) sin(2π (t) (t) = ) sin(2π Âv,lv,l v,l + Av,l (∆ sin(2π ffˆˆv,l tt + v,lf,v,l+2πt v,l+ φ̂φ̂v,l ). (12) (10) P (t) ≈ ỹP (v) (t)=∞ Av,l fˇv,l t + φ̌v,ll=0 l=0 NP (v) −1 sin(2π ) + Av,l (∆f,v,l 2πt + ∆φ,v,l ) cos(2π fˇv,l t + φ̌v,l ) l=0 (8) (t) ≈ ỹ(t) = ˇ ŷ ∆ [k] (t) · h = sr (t − Ť  [k], sin(2π {š[k], ∆ˇ fˆ t [k]}) + φ̂ + y ). + ỹ (t) + (t) + ỹ (t). (12) (9) Theclkhorizontal periodic P s (v) v,l sv,l v,l −∞ l=0 ∞ components are estimated v,R P (v) in a two-step h,rem approach as above. More k=−∞ y(t) ≈ ỹ(t) ∞ = ∆ˇ s [k] · hprecisely, l=0 sr (t − Ťclk the [k], TT sig,T [k] sig,T {š[k], estimation [k] ∆ − −based ˇ is TTsig,DD+P s [k]}) sig,DD+P +ony−∞ the(v)+ (v) [k] [k]ỹ difference time P (v) (t) + v,R (t) + ỹh,rem (t). (9) (13) (13) ˇ B. ) ≈ ỹ(t)B.=Analysis Analysis ˇ Domains ∆ sDomains k=−∞ [k] · hsr (t − Ťclk [k],[k] {š[k], (9) Tsig,T − Tĥ∆ sr s [k]}) + y sig,DD+P (v) [k] −∞ + ỹP (v) (t) + v,R (t) + ỹh,rem (t). (13) (11) k=−∞ x̂ = ŷ−∞ ĥ =A y + (10) alysis Domains where Τ sig , T [k] p̂ and Τsig , sr DD + P(v) [k] are the level crossing time of the total and the x̂ TIE=XXŷ[k] TIE [k] −∞= = T= Tsig,XA+[k] sig,X y − [k] − TTcdr cdr[k] [k] (10) (14) (14) DD + P(v) ĥ signal, sr respectively. Thus an estimate of the horizontal periodic (P(h)) contribu- tion ∞x̂to =the ŷlevel = p̂ −∞ crossing A+ ytimes is obtained and the DD + P(h) signal can(10) be synthesized. TIE [k]ˇ = T sig,X [k] − T cdr [k] ˇ (14) (11) s [k]p̂· ĥsr (t − Tcdr [k], ∆s [k]) + ŷ−∞ ŷDD (t) = X ∆∞ Summarizing, TIE TIEare we TT [k] [k]able = =to TIE TIE D[k][k] + synthesize D + TIETIE signals [k]with a subset of the underlying source(15) R[k] R (15) k=−∞ ŷDD (t) = ∆ ˇ s [k] · ĥsr (t − Tcdr [k], ∆ ˇ s [k]) + ŷ−∞ (11) Ndisturbances. ∞ P (v) −1TIE TIEDD+P DD+P(h) These [k] what-if (h) [k] = = TIE signal reconstructions TIE DD[k] DD [k] + + TIE TIEPP(h) allow us to provide new measure- (16) (h)[k] [k] (16) mentsˇ k=−∞ with selectable disturbances. ˇ For example, it is possible to construct an eye ŷDD (t) = TIE ∆ [k][k] = · TIE ĥ (t −[k] T ˆ + TIE [k], ∆ [k][k]) + ŷ (15) (11) (12)measure- (17) NTIE TIE =+TIE TIE + TIE TIE (17) ŷP (v) (t) = diagram Âwith sr sin(2π f[k]cdr v,l t= φ̂v,lDD). T s D [k] R s −∞ v,l−1 DD+P DD+P data(v) disturbances (v) and[k] DD [k] + chosen (v)[k] periodic P P(v) [k]disturbances only. The k=−∞ [k] P (v) TIEDD+P = TIE TIE TIE [k] + TIE [k] (16) ŷP (v) ments NP(t) (v) −1 l=0 = carried  (h) out DD v,l in D the = [k] [k] sin(2π D = fˆTIE TIE analysis P v,l t DD+ domains (h) DD φ̂[k] [k] + TIE + v,l ). TIE are PPdescribed [k]. [k]. in the following sections. (12) (18) (18) TIEDD+P (v) [k] = TIE DD [k] + TIE P (v) [k] (17) ŷP (v) (t) = 4.2 Analysis Âv,ll=0 sin(2π domains fˆv,l t + φ̂v,l ). (12) TIE T D [k] = TIEDD sig,T [k] − T [k] +(v) sig,DD+P TIEP [k]. [k] (13) (18) Based on the total and the synthesized signals, we move away from a signal-centric view l=0 Tsig,T LE LE− [k] X[k] [k]sig,DD+P == VVXX[k] [k] − − VVref ref[k][k] (13) (19) (19) nalysis Domains the T [k] X in order to obtain commonly(v) known jitter/noise values. Specifically, we determine TIEs defined T [k]as− T (13) B. Analysis Domains sig,T sig,DD+P (v) [k] LE X [k] = VX [k] − Vref [k] (19) alysis Domains TIEX [k] = TLE LE TT [k] sig,X [k] − LE [k] = = LE [k] + LE [k] + TcdrDD[k] LERR[k] [k] (14) (20) (20) (12) TIEX [k] = Tsig,X [k] − Tcdr [k] (14) where LE Τsig,XLE T [k] = [k]Dis[k] the+level LERcrossing [k] time of the signal X = {DD, DD + P(h), DD + P(v), D, T }. (20) TIE ForXthe TIE remaining [k][k]=T TIE[k]TIEs, − we Tcdrimpose [k] the following relationships: (14) T =sig,X D [k] + TIER [k] (15) TIEDD+P (h) [k]TIE=T [k]TIE=DDTIE [k]D+[k]TIE + PTIE (h) [k] R [k] (16) (15) TIEDD+P TIETDD+P TIE [k][k] = TIE= [k] TIE TIE [k]DD TIE + [k] + TIE[k]P (h) [k] (17) (16) =(h) D [k] + TIE R [k] (15) (v) DD P (v) TIE TIEDD+P TIE D [k] [k] = = (v) TIE=DD [k] TIE TIE [k][k] + TIE +DD TIE+ [k] TIEP (v) [k] P [k]. [k] (18) (17) (16) DD+P (h) DD P (h) TIEDD+P (v) [k]TIED [k]DD = TIE TIE =[k] +DDTIE [k]P (v) TIEP [k]. + [k] (17) (18) (13) TIELE = TIE D [k][k] VDD +VTIE [k]−analysis,P [k]. (18) (19) to refer- X for=the Similarly, [k] noise X ref [k] we determine LEs that are defined with respect ence signal levels: LEX [k] = VX [k] − Vref [k] (19) LE LEXT[k] [k]= =VLE −+ X [k][k] VrefLE [k] [k] (19) (20) (14) D R where VLE X [k] T [k] LED [k] = signal is the + LE value (20) R [k] at the symbol sampling times for the signal obtained X = {DD, DD + P(h), DD + P(v), D, T } and Vref [k] is the reference value that depends on LE T [k] = LED [k] + LER [k] (20) the current symbol value. 10
LEX [k] = VX [k] − Vref [k] (19) The remaining LEs are decomposed as in the TIE case, i.e. we use LET [k] = LED [k] + LER [k] (20) (15) TABLE I S YNTHETIC S ETTINGS and thus the remaining LEs can be computed. Property Scenario 1 Scenario 2 5 MEASUREMENTS Data rate 5 Gbps 5 Gbps Sample rate 40 GSa/s 20 GSa/s Number of symbols 300 k Variable Modulation NRZ, reference levels -1 and +1 NRZ, reference levels -1 and +1 Symbol pattern Various measurements of interest can be obtained based on theRandom Random TIEs and the LEs intro- Rate modulationduced in yection 4.2. TheTriangular TIE statistics SSC can with TABLE I either -5000ppmbe computed @ 33 kHz for Noneall values or for Step response asymmetry Fall time > rise specific subsets, e.g. rising or Sfalling time edges. YNTHETIC None Similarly, the LE statistics can be computed S ETTINGS Periodic horizontal disturbance h,P 100 mUI pp @ 20 MHz 200 mUI pp @ 20 MHz for certain symbol values only. The measurements are briefly presented in the following Periodic Propertyvertical disturbance v,P 0.1 pp @ 100 Scenario 1 MHz 0.02 pp @220 MHz Scenario Random subsections. horizontal disturbance h,R 205 mUI Data rate Gbps rms 505 mUI Gbps rms Random Sample vertical rate disturbance v,R (SNR) 3040dB GSa/s 3520dB GSa/s 5.1 Statistical values and histograms Number of symbols 300 k Variable Modulation We calculate common statistical valueslevels NRZ, reference like the minimum, -1 and +1 maximum, NRZ, peak-to-peak reference levelsvalue, -1 and +1 Symbol patternpower and standard deviation Random V. (SD) M EASUREMENTS Random for both the TIEs and the LEs. Furthermore, we use Rate modulation Triangular SSC with -5000ppm @ 33 kHz None histogram plots to illustrate the shape of their distribution. A. Statistical Values Step response and Histograms Fall time > rise time asymmetry None Periodic horizontal disturbance h,P 100 mUI pp @ 20 MHz 200 mUI pp @ 20 MHz B. Duty-Cycle and5.2Level Periodic vertical DutyDistortion cycle disturbance v,Pdistortion and 0.1 level pp @distortion 100 MHz 0.02 pp @ 20 MHz Random horizontal In thedisturbance context of 20 mUI rms a jitter analysis, h,R the duty cycle distortion (DCD) is50calculated mUI rms as Random vertical disturbance v,R (SNR) 30 dB 35 dB DCD = max {E{TIEDD |hsr,k }} − min {E{TIEDD |hsr,k }} (16) (2 k=0,...,Nsr −1 k=0,...,Nsr −1 V. M EASUREMENTS A. StatisticalLD[l] Values E {x} denotes the expectation of the random variable x and hsr,k the occurrence = and where E{LE Histograms DD |level = l + 1|} − E{LEDD |level = l}, l = 0, . . . , NPAM − 1. (2 of the k-th step response with k = 0, … , Nsr – 1. The DCD value measures the impact of B. Duty-Cycle and Level Distortion C. Autocorrelationtwo Functions and the effects. First, Power Spectral differences in the Densities various step responses, e.g. rising and falling step responses in the case of non-return-to-zero (NRZ) signals (NPAM = 2), and second, a mis- D. Symbol Error Rate Calculation match in the signal level thresholds used for determining the TIEs. DCD = max {E{TIEDD |hsr,k }} − min {E{TIEDD |hsr,k }} ( k=0,...,Nsr −1 k=0,...,Nsr −1 The level distortion (LD) for the eye opening l SER(u) = Phist [lbin ] · pR (v − xhist [lbin ]) dv (2 LD[l] = E{LEDD |level = v∈e(u) lbin l + 1|} − E{LEDD |level = l}, l = 0, . . . , NPAM − 1. (17) ( E.C. Jitter and Noise Characterization Autocorrelation Functions and Power at a Target Spectral SER Densities is used in the context of noise analysis. It characterizes vertical distortion of the eye open- D. Symbol Error ing Rate dueCalculation to data-dependent disturbances. TJ@SER = DJ + σ · c(SER) (2 5.3 Autocorrelation functions and powerδδspectral RJ densities Further information can be derived from a spectral view of the TIE and LE tracks. For F. Framework Results SER(u) = Phist [lbin ] · pR (v − xhist [lbin ]) dv ( TIEs, a spectral view is not straightforward to compute since symbol transitions do not G. Exemplary Analysis occur between every two lbin v∈e(u) symbols. Usually, this is handled by linear interpolation between E.Other Jitterstatistics: and Noise Characterization given TIEs. We proposeata different a Targetapproach SER in order to avoid the introduction of artifacts. First, compute the autocorrelation function of the TIE track by only considering the given TIE values, then compute the power spectral density (PSD) based on the autocorrelation H. Signal Length Influence TJ@SER function, thereby obtaining = DJview a spectral ( δδ + without any interpolation artifacts. σRJ · c(SER) VI. C OMPARISON AGAINST C OMPETING A LGORITHMS F. Framework Results5.4 Symbol error rate calculation VII. C ONCLUSION SER plots visualize how varying a single parameter affects the number of correctly G. Exemplary Analysis decoded symbols by plotting ACKNOWLEDGMENT the SER against the parameter in question. For a jitter anal- Other statistics:ysis, the SER is plotted against the sampling time offset used to decode the symbols, while for a noise analysis, the parameter is the decision threshold level of the considered H. Signal Length eye. Symbol errors occur, e.g. when signal transitions take place after the corresponding Influence symbol has been sampled or signal levels do not cross the threshold level of the desired symbol,VI. respectively. The SERAisGAINST C OMPARISON C OMPETING equivalent to the BER forANRZ LGORITHMS signals. VII. C ONCLUSION Rohde & Schwarz White paper | Signal model based a ecomposition 11 pproach to joint jitter and noise d ACKNOWLEDGMENT
Symbol pattern Random Random Rate modulation Triangular SSC with -5000ppm @ 33 kHz None Step response asymmetry Fall time > rise time None Normally, Periodic horizontal disturbance h,P to measure 100 mUIthe pp SER, @ 20theMHzsystem under test is200 usedmUItopp transmit a known symbol @ 20 MHz Periodic vertical disturbance sequence v,P 0.1 pp @ 100 MHz 0.02 pp @ 20 MHz and to evaluate the symbol errors in the received signal. Repeating this mea- Random horizontal disturbance h,R for a 20 surement mUI rms number 50 mUI of values of the desired parameter rmsthe SER plots. Such a yields Random vertical disturbance v,R (SNR) 30 dB 35 dB measurement setupTABLEusually I interferes with the regular operation of the system under test and may Stherefore YNTHETICbe S ETTINGS difficult to accomplish or even yield different results. Additionally, SER measurements take a very long time for low SER values like 10–12. Property Scenario 1 V. M EASUREMENTS Scenario 2 Data rate 5 Gbps 5 Gbps A. rate Sample Statistical Values and Histograms We propose 40 GSa/s a method to calculate SER plots20based GSa/s on the jitter or noise values, i.e. the Number of symbols TIEs300 or LEs, k respectively, and their statistics.Variable To calculate an SER plot, first select a histo- B. Duty-Cycle and Level Distortion Modulation NRZ, gram thatreference levels covers all -1 anddeterministic desired +1 NRZ, reference components levels with bins-1representing and +1 the desired Symbol pattern Random Random analysis domain and a set of probability distribution parameters describing the desired Rate modulation Triangular SSC with -5000ppm @ 33 kHz None Step response asymmetry DCD = random components Fall time > rise timein the {E{TIE same analysis domain. For example, when the SER for all minNone{E{TIEDD |hsr,k }} (21) max DD |hsr,k }} − Periodic horizontal disturbance h,P signal 100disturbances k=0,...,N sr −1 pp @ 20plotted mUI MHz against the sampling k=0,...,N200 timeppoffset sr −1mUI @ 20 is desired, we select the TIE MHz Periodic vertical disturbance v,P 0.1 pp @ histogram 100 MHz covering 0.02 ppand all deterministic components @ 20theMHz TIE distribution covering all ran- Random horizontal disturbance h,R dom20components. mUI rms By choosing histograms and 50 mUI rms distributions that cover only part of the Random vertical disturbance v,R (SNR) 30 dB 35 dB LD[l] = E{LEDD |levelsignal measured =l+ 1|} − E{LEwhat-if perturbations, DD |level SER= l}, can plots l =be0,generated . . . , NPAM that 1. the user(22) − allow to gain insight into how the SER would change if the user were to resolve certain signal C. Autocorrelation Functions and issues. integrity PowerThe Spectral information Densities stored in the histogram is, for every bin l bin , the prob- V. M EASUREMENTS ability of hitting that bin P hist [l bin] and the position of that bin x hist [l bin]. Assuming a D. Symbol Error Rate Calculation Statistical Values and Histogramsprobability distribution pR of the random components, we get the SER Duty-Cycle and Level Distortion SER(u) = Phist [lbin ] · pR (v − xhist [lbin ]) dv (23) (18) lbin v∈e(u) E. Jitter DCD max {E{TIEDDat|hsr,k = Characterization and Noise where e (u) = {xa k=0,...,Nsr −1 }} − SER Target : position min {E{TIEDD |hsr,k }} x generates k=0,...,N (21) sr −1 a decoding error for parameter u } holds informa- tion about the error conditions. Note that while random disturbances are often assumed to be Gaussian distributed, this may not be the case in certain scenarios due to the non- LD[l] = E{LEDD |levellinear =l+ 1|}TJ@SER transformation = DJ − E{LEfrom DD |level δδ += source l}, σtoRJ · lc(SER) = 0,domain analysis . . . , N(see − 1. 3.2). PAM section (22) (24) Autocorrelation Functions F. Framework Resultsand Power Spectral 5.5 Jitter Densities and noise characterization at a target SER Symbol G.Error Rate Calculation Exemplary Analysis A number of secondary measurements are derived from the SER calculation. One of these is TJ@SER, the total jitter at a target SER. This value represents the jitter budget Other statistics: has to anticipate when trying to reach a target SER. It can be iden- that a system designer SER(u) =tified as P the[lwidth hist bin ] · of the SER R p (vvs.−sampling x [l time hist bin ]) dvoffset plot at the desired SER. (23) H. Signal Length Influence lbin v∈e(u) Another example is the Dual-Dirac value DJδδ from the Dual-Dirac model. The Dual-Dirac VI.model Jitter and Noise Characterization C OMPARISON at a assumes thatAthe Target SER GAINST C OMPETING deterministic jitter canAonly LGORITHMS take on two discrete values and that the relationship VII. C ONCLUSION ACKNOWLEDGMENT TJ@SER = DJδδ + σRJ · c(SER) (24) (19) Framework Results holds, where the value c (SER) depends solely on the target SER. Sufficiently low SERs Exemplary Analysis need to be ensured for the above relationship to hold, otherwise the (Gaussian) random jitter does not dominate the total jitter. Other statistics: For noise analysis, we compute the EH@SER value, which gives the eye height that is Signal Length Influence achieved at a target SER. VI. C OMPARISON AGAINST C OMPETING A LGORITHMS VII. C ONCLUSION ACKNOWLEDGMENT 12
6 FRAMEWORK RESULTS In this section, we present measurement results from the joint jitter and noise analy- sis framework presented in section 4. To make the vast amount of measurements and comparisons more accessible, we present a single scenario in section 6.1 and use a majority of the available results to track down a signal integrity issue and understand its implications. In section 6.2, we show how the length of the input signal affects the mea- surements and thus the framework’s performance. The scenarios used in this subsection are summarized in Table 1. We use a second order phase-locked loop (PLL) with a band- width factor of 500 and a damping factor of 0.7 for these scenarios. Table 1: Settings for synthetic scenarios Property Scenario 1 Scenario 2 Data rate 5 Gbps 5 Gbps Sample rate 40 Gsample/s 20 Gsample/s Number of symbols 300k variable Modulation NRZ, reference levels: –1 and +1 NRZ, reference levels: –1 and +1 Symbol pattern random random triangular SSC with –5000 ppm Rate modulation none at 33 kHz Step response asymmetry fall time > rise time none Periodic horizontal disturbance ϵh, P 100 mUI pp at 20 MHz 200 mUI pp at 20 MHz Periodic vertical disturbance ϵv, P 0.1 pp at 100 MHz 0.02 pp at 20 MHz Random horizontal disturbance ϵh, R 20 mUI RMS 50 mUI RMS Random vertical disturbance ϵv, R 30 dB 35 dB (SNR) 6.1 Example analysis Throughout this section, we use scenario 1 from Table 1, i.e. a synthetic 5 Gbps NRZ data signal with random symbols to show the framework’s capabilities. Fig. 4 shows the data rate over time as recovered by the clock and data recovery (CDR). At the very start of the signal, the CDR’s locking behavior can be observed. After the lock time (indicated by the red line), the CDR follows a triangular modulation profile of –5000 ppm at 33 kHz. Such a modulation profile is commonly used for spread spectrum clocking (SSC) in, e.g. USB 3.0. Note that CDR misconfiguration happens quite frequently and is easily caught at this stage. Fig. 4: Symbol rate from CDR 1.005 Rate [nominal rate] 1.000 0.995 Symbol rate Symbol rate CDR lock time CDR lock time 0.990 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 Symbol • (105) Rohde & Schwarz White paper | Signal model based a ecomposition 13 pproach to joint jitter and noise d
The recovered clock can be used to produce eye diagrams not only of the total signal, but also of synthesized signals covering only a subset of the disturbances in the total signal. This allows the user to graphically assess the most severe causes of signal degra- dation and to acquire an expectation of what could be achieved in a given scenario once a certain issue is fixed. The variety of available synthetic eye diagrams improves upon the possibilities offered by state-of-the-art methods. Fig. 5 shows some of these eye dia- grams: a) the total signal, b) the data-dependent signal, c) the data-dependent signal with periodic horizontal disturbances and d) the signal with all deterministic components. We can see that the data-dependent component is responsible for most of the jitter and noise. Successively adding the periodic disturbances yields signals that approach the jit- ter and noise of the total signal. Fig. 5: Total and synthetic eye diagrams a) T b) DD c) DD+P(h) d) DD+P(h)+P(v) Deeper understanding of the individual jitter and noise components is obtained from TIE and LE histograms as in Fig. 6. These show the distribution of all timing and level errors for various components. Some effects can be seen more easily using TIE histograms for rising and falling edges and LE histograms for symbols 0 and 1. Subplot c) in Fig. 6 shows that falling edges cause significantly more data-dependent jitter than their rising counterparts. Periodic jitter in e) adds a moderate amount of additional jitter. In the noise domain, data-dependent components dominate over periodic components. The random components in g) and h) show a close to Gaussian distribution. 14
Fig. 6: TIE and LE histograms a) TIET b) LET Frequency Frequency Rising Symbol 0 2000 Falling 4000 Symbol 1 1000 200 0 0 –0.2 –0.1 0 0.1 0.2 –0.4 –0.2 0 0.2 0.4 TIE [UI] LE Frequency c) TIEDD d) LEDD Frequency 8000 Rising Symbol 0 8000 Falling 6 Symbol 1 6000 4 4000 2 2000 0 0 –0.2 –0.1 0 0.1 0.2 –0.4 –0.2 0 0.2 0.4 TIE [UI] LE e) TIEP f) LEP Frequency Frequency 4000 2000 Rising Symbol 0 Falling Symbol 1 3000 1000 2000 1000 0 0 –0.2 –0.1 0 0.1 0.2 –0.4 –0.2 0 0.2 0.4 TIE [UI] LE g) TIER h) LER Frequency Frequency Rising 6000 Symbol 0 3000 Falling Symbol 1 4000 2000 1000 2000 0 0 –0.2 –0.1 0 0.1 0.2 –0.4 –0.2 0 0.2 0.4 TIE [UI] LE Rohde & Schwarz White paper | Signal model based a ecomposition 15 pproach to joint jitter and noise d
You can also read