DAC1 Designer Talks about Jitter, Wi-FI And
The Future of 32-Bit Digital Audio Converters
Editor's Note:
Benchmark Media Systems, Inc. was founded by Allen H. Burdick in 1985. Benchmark began as a manufacturer of high-performance audio products for television network facilities. Burdick's innovative designs pushed audio technology well beyond what was commonly available in 1985, especially for TV audio. Benchmark’s product line soon expanded to include microphone preamplifiers, headphone amplifiers, and meter systems. Benchmark’s microphone preamplifier soon attracted the attention of recording engineers and Benchmark expanded its market to include recording studios. The products were also of such quality that Burdick, an avid audiophile, enjoyed the performance of the broadcast-oriented Benchmark DA101 distribution amplifier (bandwidth of 0.1 Hz to 160 kHz, a SNR of more than 130 dB, and THD+N of 0.00044%) in his own home.
In 1995, John Siau entered the Benchmark world, first as a consultant. Burdick had decided to add digital audio products in the tradition of Benchmark’s analog product quality, which fully immersed the company in the recording studio as well as broadcast. I came to know Benchmark and Siau in 1995, as I launched the magazine,Pro Audio Review.
John Siau moved to full-time design duties and fit in perfectly at engineer-driven Benchmark Media. He brought 15 years experience designing A/D and D/A converters, sync circuits, and digital signal processing circuits for the fledgling high-definition television technology. His designs included low-jitter oscillators and PLL circuits to meet the demands of high-definition video projectors. At Benchmark, Siau's first design was the AD2004, a 20-bit 4-channel A/D converter.
By 2001, Siau guided Benchmark into a new generation of digital products that fit both the professional and audiophile niches, including the well-regarded DAC1, DAC1 Pre, ADC 1 A/D, DAC1 HDR and the extraordinarily transparent MP1A stereo microphone preamp.After Burdick retired in 2007 for health reasons, Siau took over Benchmark’s operations. He also maintains his position as the company’s primary design engineer. He is quick to point out that in this day of overseas production, all Benchmark products are designed and manufactured in Syracuse, New York.
I recently conducted a Q/A session with Siau to find out more about his design philosophy and the future course for Benchmark Media.
—John Gatski
EAN: What is your background in audio engineering? Did you to go school to specialize in audio?
Siau: I grew up with a basement full of electronic parts, old radios, and HAM equipment. From the time I was 8 or 9 I knew I wanted to design electronic products. In 1976, I enrolled in the engineering program at Syracuse University. Shortly after beginning my studies I took on a side job repairing a sound system for a local band. I subsequently became their sound engineer. About that time, I began piecing together my own personal audiophile system. I graduated in 1980 and took my first job designing video equipment for the CBS Technology Center in Stamford, CT. Five years later I moved my family to Syracuse and continued in video design for GE for another 10 years. Throughout that adventure I continued enjoying audiophile systems, and I continued mixing live sound.
EAN: How did you end up working for Benchmark Media?
Siau: I was originally hired as a consultant in 1995 to design Benchmark’s first digital product — the AD2004 20-bit A/D converter. I am proud to say that it set new “benchmarks” for low THD+N and low jitter, and quickly won industry awards.
EAN: What are some of the Benchmark products you’ve had a hand in designing over these 15 years?
Siau: My designs include the AD2004, DAC2004, AD2404, AD2408, DAC2404, DAC2408, ADA2408, DAC104, ADC104, DAC1 family, HPA2, ADC1 family, MPA1, PRE420, and three OEM conversion systems. In addition to these, I transferred Benchmark’s older analog designs from through-hole to surface-mount technology.
EAN: Benchmark is a company that builds audio products for the professional and consumer/audiophile markets. Does that engineering convergence ultimately benefit the pro or the audiophile end-user, or both?
Siau: The audiophile certainly benefits from the performance and quality of true professional-grade products. The professional user benefits from the R&D dollars generated by audiophile sales. This mutual benefit only persists when the manufacturer is committed to maintaining high performance and high build-quality. Benchmark will not compromise performance or quality to satisfy the demands of the consumer market. Our products are intentionally positioned outside of the mass-produced consumer audio market. Our sales volumes are high enough to make our products affordable, but not high enough to attract offshore clones.
The 32-bit systems will offer no advantage until converters reach a SNR of about 138 dB, but many other things would have to change as well. The best line-level analog circuits barely exceed 130 dB, many power amplifiers barely exceed 16-bit (96 dB) performance.
EAN: One of your most successful consumer/pro product lines has been the DAC1 D/A Series. From the beginning, Benchmark has touted its sonic advantages, especially its jitter reduction. Jitter is often misunderstood. What is jitter, and why is it so important to eliminate it?
Siau: Jitter is the time variation of a periodic signal. Jitter on an audio sampling clock can cause non-harmonic non-musical distortion. This distortion can be much more audible than harmonic distortion. Jitter has been well understood by video and communications engineers for many years, and there are many well-established techniques for attenuating and controlling jitter.
Unfortunately, jitter was often ignored in the early days of digital audio. Many engineers assumed that jitter was unimportant in audio conversion systems. I would like to think that those days are over, but I still see “high-resolution” audio A/D and D/A converters that are clocked directly from AES, S/PDIF, or USB receiver chips.
EAN: So jitter is distortion?
Siau: Any jitter on an A/D or D/A conversion clock will add non-harmonic distortion that is very unmusical and unnatural. This distortion can be reduced to inaudible levels if conversion clock jitter is low enough. There are several jitter-attenuation technologies that can keep jitter-induced distortion at levels that are -130 dB to -150 dB relative to the level of the music. At these levels, jitter-induced distortion is well below audibility and even well below the threshold of hearing (at any sane playback level). We are one of several manufacturers that deliver products that keep jitter-induced sidebands at least 130 dB below the audio.
Unfortunately, jitter attenuation circuits cost money and can be difficult to implement properly. For one or both of these reasons, many conversion systems produce jitter-induced distortion that is only 60 to 80 dB below the audio. In these systems, jitter-induced distortion can easily exceed the threshold of hearing at common playback levels.
EAN: What are the audible effects of jitter?
Siau: The initial impression is usually a change in frequency response. Two systems with ruler-flat frequency response can sound like they have very different frequency responses. High-jitter systems often sound like they have a fuller midrange. Low-jitter systems often sound like the bass and highs are slightly boosted or extended. I suspect the reason for this is that jitter-induced distortion tends to fill and clutter the midrange. Mid-frequency musical details are often obscured by jitter-induced distortion. After the initial “frequency response” impression wears off, low-jitter systems seem to reveal more midrange detail.
By the way, jitter cannot alter or blur L/R positions within the stereo image. In most conversion systems, the left and right channels share a common sampling clock. Sampling clock jitter moves both channels together in time and consequently there is no L/R movement in the stereo image.
EAN: So if I care about jitter degrading my performance, I should look for a digital converter with good jitter reduction results? What is the optimum measurement spec?
Siau: The best test is an FFT analysis of a high-amplitude 10 kHz test tone carried on a high-jitter digital signal. Jitter will produce unwanted “sideband” energy above and below the 10 kHz tone. The results of the FFT analysis can be summarized in a specification for “maximum jitter-induced sideband amplitude” or “maximum distortion due to jitter.” These measurements tell the whole story, and they indicate what can be expected in the real world. In contrast, specifications for “jitter amplitude” are very misleading and nearly useless. The audio industry is full of misleading and erroneous jitter measurements.
Common audio measurements, such as THD+N and SNR, usually fail to expose a jitter problem. The THD+N bench tests are normally performed under very ideal conditions, using low-jitter digital-audio test signals. In real-world applications THD+N may rise significantly. SNR (and dynamic range) tests fail for a different reason: jitter does not normally produce noise or distortion unless an audio signal is present. Converters with excellent THD+N and SNR specifications may still have significant jitter problems.
Most A/D and D/A converter chips perform better at 96 kHz than at 192 kHz. There are now a few chips that perform equally well at 192 kHz. Nevertheless, my position is unchanged. I believe 96 kHz is more than sufficient. The brick wall filter in a 96 kHz system is centered at 48 kHz and has no significant impact on audible frequencies.
EAN: What about interface jitter? Is there one digital interface that performs better than another?
Siau: In a well-designed system, sampling clocks are well isolated from interface clocks, and interface jitter has no significant impact on performance. Interface jitter can cause sampling clock jitter in a poor-quality converter. Jitter on interfaces such as AES, S/PDIF, ADAT, USB, and FireWire are only an issue when a converter lacks sufficient jitter attenuation. The highest speed USB and FireWire interfaces have jitter that is intentionally added — in an effort to reduce radio interference. Again, this interface jitter is inconsequential in a well-designed system. Near-perfect jitter performance can be achieved when a manufacturer is willing to take the necessary steps to attenuate and isolate jitter.
EAN: Your converters, such as the DAC1 series, utilize asynchronous conversion. Explain to the reader how asynchronous conversion works. Why do so many converter companies implement this method of conversion in audio DAC?
Siau: Asynchronous sample rate conversion (ASRC) is a digital-to-digital filtering process that changes the sample rate of a digital signal. ASRC circuits can be used to frequency-shift the digital filters in a D/A converter IC so that these filters can be replaced with a better digital filter (to reduce aliasing or imaging). ASRC can also be used to isolate a high-jitter data-transfer clock from a low-jitter data conversion clock. Any jitter isolation provided by the ASRC is only effective if the input-to-output sample rate calculation is filtered to reject jitter.
Unfortunately, most ASRC devices have little or no jitter attenuation. However, there are a few ASRC devices that have extraordinary jitter attenuation. For example, the Analog Devices AD1896 has over 100 dB jitter attenuation at a jitter-frequency of 1 kHz, and its jitter attenuation extends down to 2 Hz. The AD1896 can be a very effective component of a jitter attenuation system. There is a “pin-compatible” replacement for the AD1896 that claims higher performance. Unfortunately, this “upgrade” ASRC has no jitter attenuation.
EAN: What are the disadvantages of asynchronous conversion?
Siau: The ASRC process is computationally intensive. Millions of high-precision multiply and accumulate operations must be performed every second. The accuracy of the process is only limited by the amount of processing horsepower we are willing to apply. The best ASRC devices have high-precision DSP engines and achieve THD+N performance that is better than -130 to -150 dB. On the basis of performance, a high-quality ASRC can be justified if the distortion it introduces is far less than the jitter-induced distortion that it prevents. There are a few ASRC devices that easily achieve this goal, but the majority lack adequate jitter-attenuation.
EAN: With 24-bit converters now measuring more than of -125 dB in signal to noise and dynamic range — offering audible improvement over 16-bit.—rewrite for clairity. Will moving to 32-bit converters produce any further audible benefits? Or have we maxed out with 24-bit, as far as improving the sound by increasing the word length?
Siau: The 32-bit systems will offer no advantage until converters reach a SNR of about 138 dB, but many other things would have to change as well. The best line-level analog circuits barely exceed 130 dB, many power amplifiers barely exceed 16-bit (96 dB) performance, and today’s best 24-bit recordings barely exceed the SNR of a 16-bit system. Most audiophile systems are limited by the SNR of the volume control circuits. Most power amplifiers do not come anywhere close to delivering 24-bit performance (144 dB).
One of the goals of the DAC1 was to simplify the audiophile signal chain. The DAC1 is designed to directly drive power amplifiers. In such a system, the power amplifier will be the limiting factor. We need better power amplifiers before we will need 32-bit converters.
EAN: Numerous digital converter modifications are on the market. Can we audibly gain any more from the converter by implementing better performing analog parts in the signal path?
Siau: Beware--this is a dangerous path! We have seen aftermarket modifications to our DAC1 converters. Every single one of these units has had lower performance than the factory units. Aftermarket replacements of the op-amps and of the ASRC have proven particularly destructive to the performance. I have no doubts that the modified units sound different than stock units, but, in most cases, it is not for the better.
EAN: A few years ago when we were talking about the DACs, you told me that it is not necessary for 24-bit converters to go beyond the 96 kHz sampling rate. You said that increasing the sample rate does not produce better subjective audio performance or better measured specs. And in the case of 192 kHz, the higher rate can reduce measured noise performance. Does sampling at 176 kHz or 192 kHz adversely affect the sound quality, or is it just a measurement negative?
Siau: Most A/D and D/A converter chips perform better at 96 kHz than at 192 kHz. There are now a few chips that perform equally well at 192 kHz. Nevertheless, my position is unchanged. I believe 96 kHz is more than sufficient. The brick wall filter in a 96 kHz system is centered at 48 kHz and has no significant impact on audible frequencies. In contrast, the filters in 44.1 kHz and 48 kHz systems can cause problems. Our equipment supports 192 kHz, but we recommend 96 kHz for critical applications. The differences between 96 kHz and 192 kHz may not be audible, but 96 kHz will usually measure better than 192 kHz.
EAN: Your converters have always contained the full array of traditional connection options, such as SPDIF (RCA), TOSlink (optical) and AES/EBU XLR. With the USB series of converters, you allowed computer users an easy path to your DACs? Is USB as transparent as SPDIF or XLR in terms of audio quality? Or is it just a necessary “evil” to get a common digital audio link to the computer?
Siau: USB is only transparent when it supports 24-bit audio. The 16-bit USB interfaces are worthless — as they almost always cause truncation – even when playing 16-bit files. Computer playback systems have digital volume controls that expand audio word-lengths to 24-bits. If the playback hardware cannot support 24-bits, the lower bits are truncated (without dither), causing audible distortion. USB products must be capable of responding to sample-rate change requests. If a client cannot respond to such a request, the operating system will “fix” the problem by inserting low-quality sample rate conversion. Unfortunately this OS-inflicted SRC is of very poor quality, and the distortion is very audible.
And USB products must have sufficient jitter attenuation to remove the relatively high levels of jitter on the USB interface. USB specifications actually mandate that jitter be applied to reduce radio interference. But the good converters, such as the Benchmark DAC1 series, easily remove this jitter. Converters with sufficient jitter attenuation will produce identical results from all common digital audio interfaces, provided the interfaces are bit-transparent.
It may surprise you that I have measured significant differences in speaker cables. People have claimed audible differences in cables, and it turns out that differences can be measured! BTW, zip-cord isn’t half bad! But maybe this is a topic for another day.
EAN: With Blu-ray audio emerging (linear PCM up to 24/192 and lossless high-res from lossless Dolby Tru-HD and DTS Master HD in 7.1 channels), does Benchmark have any plans to adopt HDMI connection for its converters?
Siau: We are always looking at all available interfaces for digital audio. HDMI is currently in a state of flux, but its adoption by some computer manufacturers is encouraging. Ultimately, the computer industry will determine the most popular interface for digital audio. Other contenders are USB, FireWire, and Ethernet AVB. All are capable of transporting audio transparently — if jitter is adequately attenuated.
EAN: What about Wi-Fi? High-end Wi-Fi converters have already hit the market. What about a Benchmark converter via Wi-Fi?
Siau: We recommend using Wi-Fi client devices such as the Squeezebox for transparent Wi-Fi connectivity and full control of music playback. These mass-produced devices are so inexpensive that it is hard to justify building them into a product like ours. These devices will become obsolete long before Benchmark owners retire their DAC1 converters.
EAN: The audio world is often filled with tweaks and gizmos that claim to improve the sound of audio; many of those improvements don’t show up in test measurements? Are there some audible differences that just can’t be measured? Or do you believe that the claimed improvements are not really there?
Siau: I firmly believe that we can measure many defects that are not audible. I am sure that there are a few audible defects that we are not detecting with our test equipment, but I believe these are rare. It may surprise you that I have measured significant differences in speaker cables. People have claimed audible differences in cables, and it turns out that differences can be measured! BTW, zip-cord isn’t half bad! But maybe this is a topic for another day.
EAN: Shifting gears for the moment, tell me about the new video series, "Masters of Their Day," that Benchmark has produced. I understand the first one is already available with high-res music download capability. What prompted you to make these videos?
Siau: Our goal was to create a conversation about recording techniques, equipment, and methods. The videos bring the viewer into the studio to experience the process of creating a good recording. To make things interesting, we added a time constraint – one 8-hour day to record and mix a single. Episode 1 actually took place in less than 5 hours due to a blizzard in NY
We have CD-quality and high-resolution mastered versions of the mix available for free download, and very shortly we will have individual tracks. Visitors to the dedicated website, www.mastersfromtheirday.com, will be able to download tracks and listen to individual instruments recorded with Benchmark preamplifiers and converters. Microphone selection and placement can be viewed on the video as well. Benchmark’s Elias Gwinn, producer of the series, will be available to answer questions that are posted to the forum. We may eventually post all original tracks and challenge visitors to create their own mix and post it back to the site.
EAN: Thanks for talking with me at The Everything Audio Network.
Siau: Thank you.
No comments:
Post a Comment