Picosecond Time Measurement

PicoQuant offers several instruments that can be used to perform time interval analysis on time scales ranging from picosecond to seconds. All units require an electrical input signal, which can, e.g., be obtained from a single photon sensitive detector with a corresponding high temporal resolution.

Time-tagging units

HydraHarp 400

Multichannel Picosecond Event Timer

• Up to 8 independent input channels and common sync channel (up to 150 MHz)
• Time channel width of 1 ps
• Time tagging with sustained count rates up to 40 Mcps
• USB 3.0 connection

PicoHarp 300

Compact Dual-Channel Picosecond Event Timer

• Two identical synchronized but independent input channels
• Time channel width of 4 ps
• Time tagging with sustained count rates up to 5 Mcps
• USB 2.0 connection

TimeHarp 260

Dual- or Triple Channel Picosecond Event Timer

• One or two independent input channels and common sync channel (up to 84 MHz)
• Two models with either 25 ps (PICO model) or 250 ps (NANO model) base resolution
• Ultra short dead time (< 25 ns for PICO model, < 2 ns for NANO model)
• PCIe interface

Picosecond Time Measurement in Digital Audio Signals

Introduction

PicoQuant's Time-Correlated Photon Counting instruments of the TimeHarp, PicoHarp, and HydraHarp family are essentially high resolution time measurement systems that can also be used as generic Time Interval Analyzers (TIA) with picosecond resolution. Applications for such precise time measurements exist in many areas of design, calibration, automated testing, and quality control of high speed digital circuits and systems. Even though modern oscilloscopes can cover some of these applications, the specific requirements in automated routine testing such as throughput, automation interfaces, and cost often call for different solutions. This application note demonstrates the capabilities of PicoQuant's TIAs in a conceptually relatively simple measurement task in digital audio transmission.

It is well known to audio enthusiasts and audio engineers that the sound quality of digital audio not only depends on sufficient amplitude resolution and sampling rate but also on the precise timing of each sample at the moment of A/D and D/A conversion. In typical consumer grade digital audio systems it is common practice to transmit audio signals via S/PDIF (Sony/Philips Digital Interface Format). The signal is transmitted either electrically through coaxial cables or through fiber optic cables. Optical transmission has the benefit of better immunity to interference from electromagnetic noise but ultimately requires conversion to and from electrical signals at either end. In any case, a key feature of S/PDIF and related formats is that data and clock are transmitted through the same serial line. This is a strength in terms of simplicity, reliability, and cost but can be a weakness in terms of accurate clock transmission.

S/PDIF and related formats employ a biphase-mark-code that has either one or two transitions for every bit. This ensures that from an electrical point of view one deals only with AC signals so that DC offsets and amplitude variations do not matter much. The code also carries special block preambles marking a word boundary, thereby allowing the original word clock to be extracted from the serial signal. If the extracted clock is directly used at the receiver to clock the D/A conversion, it is obvious that sound quality will depend on the quality of the transmitted clock and any degradations the transmission may incur. While the clock frequency stability depends primarily on the quality of the transmitter's clock crystal, there can be significant timing degradation due to jitter in the recovered clock at the receiver.

Jitter describes the short term deviation from true periodicity of the clock signal and may also be interpreted as phase noise. It is to some extent present already in the clock source but in the case of transmission over a cable it is usually increased by noise pickup. In the case of S/PDIF transmission things are further complicated by effects of the transmitted data on the receiver. Since the signals are AC coupled, the long term integral must be zero. The AC coupling requires capacitors or RF transformers that have certain time constants which determine the duration of the integral, i.e. their „memory“. This memory is inevitably longer than a few bit periods which means that the content of the previously transmitted data has an influence on the state of the receiver's input circuitry while it is detecting a new signal transition. This may lead to minute timing shifts in detecting a transition. Since the transmitted data is more or less arbitrary, it creates an effectively signal correlated but hard-to-predict jitter in the recovered clock.

While early receiver circuits in the 1980s had very little protection against such clock jitter, it is today quite common to improve the clock recovery by means of phased locked loops (PLLs) and re-clocking of the received data before feeding it to the D/A converters. Nevertheless, such PLLs can only reduce but not entirely eliminate clock jitter. It is therefore still of interest to try and deliver signals with low inherent jitter and steep signal edges so that optimal clock recovery is made possible.

Measurements

In the following we show time domain jitter measurements on the S/PDIF output signals of three different audio interfaces for playing digital music from a computer through a DAC + amplifier or a home theater system with S/PDIF input. In all cases we use a HydraHarp 400 for the timing measurements. The HydraHarp 400 has up to 9 timing inputs of which we use two, one for the rising and one for the falling edges of the S/PDIF signal. The HydraHarp 400 has a digital time resolution of 1 ps and a timing uncertainty of about 10 ps r.m.s., which will determine the limit to which we can make accurate jitter measurements. In principle the HydraHarp 400 can record all timing events to a file for detailed analysis, however, for simplicity we only use its real-time histogramming mode here. In histogramming mode all timing events (signal transitions in this appnote) are timed against a common sync channel and the timings are put in a histogram for immediate display.

By means of a small signal adaptor we configure the system such that all rising edges are triggering the sync channel of the HydraHarp 400 and all falling edges are triggering an input channel of the device. We will therefore collect a histogram of time differences between falling and rising edges in the S/PDIF signal. Since the biphase-mark-code has three different durations of low/high periods we obtain a histogram with three peaks. The widths of these peaks are a direct indication of the jitter in the S/PDIF signal. By histogramming over periods of one second we obtain sufficient statistics and also cover a sufficiently long time span to include low frequency jitter components. This measurement scheme works for all common sampling rates supported by S/PDIF but for brevity we only show data collected at a sampling rate of 44.1 kHz (CD quality).

The first figure shows a screenshot of the HydraHarp software with the timing histogram from the S/PDIF output signal of a Realtek ALC899 codec chip integrated on a PC mainboard. The histogram (log scale) looks rather ragged and actually shows double peaks with a spacing of a few ns between them where there should be only one. Observe the statistics at the bottom of the histogram display: The software performs a real-time analysis of the data and shows the full width half maximum (FWHM) of the largest histogram peak (1.488 ns here)*. Considering the dual peaks we are effectively seeing a timing jitter on the order of several nanoseconds, which is actually in agreement with the specifications of the ALC899. Unless the receiver provided a very good clock recovery or re-clocking, such levels of jitter would be clearly detectable in spectral analysis and likely be audible.

Having mentioned the effect of data on clock jitter due to AC coupling, it is interesting to look at a signal of variable content versus one of fixed content. Figure 2 shows a screenshot of a measurement on the same mainboard output where instead of music only „silence“ is played, i.e. just digital zeros. Due to the exactly repeating signal structure the FWHM figure is now down to 1.1 ns, thereby confirming the effect of data induced jitter. The bi-modal peak structure is still there, probably due to some ubiquitous electrical noise on the computer mainboard or some weakness of the codec design.

For comparison, a low cost Douk Audio USB-to-S/PDIF interface was put to test next. It is not entirely clear who the original manufacturer is because it is sold on ebay under different names. Nevertheless it can be confirmed that it uses an XMOS U8 USB audio class 2 interface that should outperform the mainboard codec. Figure 3 shows a screenshot of the jitter measurement while playing music. Indeed, the histogram is now showing clean peaks and the FWHM jitter figure (0.192 ns) is dramatically improved over the mainboard output. Repeating the test with „silence“ we obtain Figure 4, confirming the effect of data on jitter one more time.

The last device put to test for this small application study was a M2Tech Hiface Two, also a USB-to-S/PDIF interface employing an XMOS USB chip. It is advertised with particular emphasis on its low jitter and sold at a price indicative of a higher end market. Indeed, Figure 5 shows clean histogram peaks and an even lower FWHM jitter value of only 64 ps. Finally we also performed the „silence“ test with this device (Figure 6) and again it can be demonstrated that the jitter goes further down (FWHM 32 ps), which is mostly of academic interest but shows that we had not yet reached the resolution limit in our measurement with music.

Conclusions

We hope to have shown how PicoQuant's timing devices can be used in generic electronic metrology applications, using the HydraHarp 400 and digital audio signals as an example. The same measurements could have been performed with the PicoHarp 300 and TimeHarp 260 devices that provide slightly lower time resolution but some cost benefits that may be of interest in certain application scenarios.

It should be noted that the jitter measurement shown here was only a very basic demonstration. The time tagged event recording of the TimeHarp, PicoHarp, and HydraHarp family would allow for a much more detailed analysis of the data. In fact, it would be possible to measure jitter between word boundaries, frames, or across any arbitrary time span. Even the entire digital audio file could be reconstructed from the event data and then be compared against the source file. The programming effort to achieve this is only moderate but was not justified for the purpose of this simple study.

Finally we would like to point out that the objective of this study was a conceptual demonstration and not an evaluation or endorsement of the various audio interfaces put to test. Audio jitter and its degree of audibility is a much debated topic to which we do not wish to contribute more than some measurement results can say. Enjoy the music, anyway.

* Note that jitter specifications are usually given as r.m.s. values. In case of a Gaussian distribution the r.m.s deviation corresponds to the standard deviation σ and FWHM ≈ 2,35 σ.