A DIGITAL CONVERTER in touch with your emotions
Multibit synchronous DAC with vintage PCM56
<< Ready for network streamers and "liquid music"
it was time to build my first DAC from scratch!
Why build it and not simply use the converter of commercial players?
And why design it with "outdated" chips?
The DAC chip is certainly important,
but the biggest contributor to the way a digital converter plays
is the analog output stage, which primarily determines the DAC sound signature.
Here, the AD826 Analog Device opamp, in conjunction with the impressive AD828,
reconstruct voices and atmospheres in such a vivid way
hard to find in commercial players.
In my listening tests,
the old PCM56, a Burr Brown multibit converter well regarded in late '80s,
sounded competitive and more natural than modern DACs,
especially when running in parallel and without oversampling.
On top, fast comparators and a local PLL clock board recover the digital signals
to significantly cut down "jitter" during the conversion process>>
Copyright (c) by Gianluca G, Italy 2019
This is an independent hobbyst project and non-commercial site: no SEMICONDUCTOR MANUFACTURER has any association with this work.
The information here presented is believed to be technically correct and everything presented on this site is done so in good faith. Anyhow you (the reader) are responsible for anything that you might do as a result of reading here.
This project took me one entire year from conception and research to realization.
Nearly all parts of this DAC has been simulated by me, before contruction, on LTSpice, a free high performance SPICE simulation software available on Analog Device website.
Gianluca G., Italy 2019
The 100% homemade audio chain will be soon enhanced by a DIY power line conditioner
"Liquid music" refers to all pieces of music available as computer files. It is music no more linked to any specific support (CD, DVD, vinyl, tape), which was first ripped in digital form - either entirely (WAV) or compressed without loss of information (FLAC) - and then stored in mass storage devices such as network server, hard disk or memory card.
Therefore, "liquid music" can be easily transported from one device to another without loss in quality and degradation over time, and it is stored in enormous quantity to be played by multiple devices in variegate places.
Since amplifiers, speakers and our ears do not process digital information but only continuous analog waveform, any digital player requires a digital-to-analog converter (DAC) to re-convert the digital content of the music into the original analog audio signal to be amplified and reproduced by the speakers.
Virtually all commercial digital players include a DAC to generate the analog audio output.
The ordinary audio conversion at 16 bits/44.1kHz, which is definitely a good enough "CD quality" resolution, is a fairly mature technology and most of players on the market today are well designed and have perfectly adequate DAC inside. However, for many players you can bypass the internal DAC and use them simply as digital sources for an independent external DAC, which is supposed to be of a superior quality compared to the inner one.
Of course, the better is the DAC, the truer to the original recorded signal will be the analog output, and the more beautiful will be the sound to our ears: differences among DACs are subtle, but they do exist.
Every converter needs a digital clock signal, which is highly critical during the conversion: the stability and the noise of the clock affects the audio output by telling the DAC chip "when" to convert.
If the sampling frequency is 44.1 kHz (44,100 samples per second), the time interval between each sample is only 22.7 μs (microseconds): in a 16bits/44.1kHz digital recording system the analog music signal is converted every 22.7 microseconds into a single digital value out of 65536 possible values (=2^16 i.e. 16 binary weighted bits). The better is the DAC timing system, the more accurate will be the re-conversion from binary values into analog sound: every small fluctuation of the time the DAC processes the digital data, recreates a distorted analog wave with imperfect amplitude and frequency; it's as if the DAC lost its focus.
The major weakness of commercial digital players are the stability of the internal clock (which is, in the best case, only a quartz plus a pair of capacitors) and the noisy power supply that feeds it (which is, in general, a single cheap voltage regulator which is shared by various analog and digital DAC sections). In same DACs circuit, crystal is even missing!
If you go for a standalone DAC, it should have a much more accurate and stable clock than the player's one. This should be powered by a separate and exclusive power supply circuitry, filtered by cascade low noise voltage regulators to keep external noise down, which otherwise would influence the oscillation of the clock.
The DAC clock should also be "in sync" with the digital streaming source: that means it should recover and realign with extremely precision ("re-clock") all the edges of the digital signals running inside the DAC machine, matching with the external source the input protocol rate of the internal DAC chip.
Another good reason to buy, or better, to build an external DAC is given by the complete freedom in the design of the analog output stage, which can be tailored ad hoc to suit your personal taste, your genre preferences and the needs of your amplifier and speakers.
De facto, the biggest contributor to the way a digital player sounds can be found in the analog output stage which primarily determines the DAC sound signature.
Also in this case, the power supply must be a perfect stable voltage source to minimize noise contamination and let the operational amplifier output be governed exclusively by its inputs. Better when this power supply circuit is disconnected from the others, especially from those feeding the digital sections. Even better, if built it in a dual-mono mirrored configuration to eliminate interference between the right and left audio channels.
The analog stage includes a separate low noise regulated power supply board and a bank of Paper-Oil & Teflon capacitors at its output.
"The 51st Anniversary DAC" power supply is split over five separate boards and uses four 25W toroidal transformers with separate windings, 28 turboswitch ultra-fast soft recovery diodes, 14 voltage regulators (mainly complimentary pair, LD, ultra low-noise LT1963/LT3015), 2 local voltage regulators based on low-noise BC550C, and a total capacitance of 140.000uF (+135°C grade).
The 51st Anniversary DAC power supply is split over five boards with separate transformers, bridges and regulators, in order to prevent the noisy digital power supply from polluting the clean analog one. The total capacitance is 140.000 microfarads
Almost all commercial DACs use "oversampling technique": they increase and multiply (X4, X8 etc..) the original sampling frequency of the digitally recorded signal by generating news samples in between every two adjacent.
In order to understand oversampling advantages, we must understand the Nyquist-Shannon theorem (1949) that establishes the minimum number of measurements needed - without introducing errors - to convert a continuous analog signal into a binary digital sequence. The minimum frequency of digital measurements (or sampling rate), required to achieve a lossless reconstruction of the original analog signal, must be greater than twice the frequency of the signal: this explains why lossless digital music is sampled at a minimum of 44.1KHz (or 44,100 samples per second), which is just above twice the limit of human hearing (20KHz).
In general, the analog signal at the output of the converter must be cleaned (i.e. low-pass filtered) over half of the sampling frequency to avoid that spectral copies of the signal and unwanted high frequency noise arrive at the amplifier and speakers creating problems like intermodulation, power wasting, ringing or oscillation, and, even, speaker damaging.
Hence, in the case we use a 44.1KHz sampling rate, we should reject anything over 22.05KHz, but nothing before 20KHz: this requires high-slope analog filter characterized by strong phase distortion and audible "group delay". "Group delay" is a measure of the speed at which signals of different frequencies propagate through the filter and it is always bad: since you will receive some frequencies before others, the sound image will lose focus and clarity, so the music will get coloration and artifacts.
Oversampling was developed to allow a simpler and less critical analog filter: oversampling performs low pass filtering of the ultrasonic image bands in the digital domain, lightening the requirements of the analog filter, which can act in a wider transition band.
Oversampled DAC has also the advantage of suffering less from the response attenuation caused by the recontruction process (function of sin(x)/x envelope): a 2X oversampling produces a drop of one decibel at 20KHz compared to more than -3dB at the output of a DAC without oversampling. So, in order to improve the tonal balance, if we use a N-OS (non-oversampling) DAC we will need to correct this attenuation with a peak in the analog filter to gently boost the two highest audible octaves (from 5KHz to 20KHz), making further critical its design.
Oversampling or non-oversampling is a very controversial subject among listeners. After comparative tests, I preferred to build a N-OS DAC, without extra chips for oversampling and digital filter: many people, like me, prefer to leave the digital signal unchanged, enjoying something softer than a "brickwall" digital filter, with less ringing and artifacts. A well designed N-OS can offer a more natural and musical sound in terms of timbre, as well as a better focused image with a deeper soundstage: modern oversampling (including Delta-Sigma converters, with their arguable feedback digital loop) initially can sound exciting, detailed and seductive to the ears, but ultimately annoying and, sometimes, even fatiguing. At the end I would say: "If you enjoy listening to music, you should get a N-OS multibit; if you like more musical details, then go for oversampled and Delta-Sigma converters."
The "51st anniversary" design relies on a Class A op-amp analog output stage which is split into two main sections: one for the conversion of the DAC output current into voltage and the other for the low pass filtering and output cable driving.
The overall filter slightly boosts (+1.3dB) the last two audible octaves to partially compensate the frequency attenuation caused by the recontruction process (not included in this simulation). At the end, the measured drop at 20KHz results to be only -1.4dB.
The filter is designed for a non-audible phase distortion due to the minimization of the group delay (+4.6µs), as I found myself more sensitive to phase distortion than to intermodulation distortion. That resulted, in the simulation, in about -6dB of attenuation of the unwanted spectrum at 44.1KHz (and -20dB at 88.2KHz and -35dB at 176.4KHz).
Such a filter appears on the paper totally inadequate as it leaves unfiltered a large part of the ultrasonic sampling images that occur at multiples of the sampling frequency.
However, I discovered that this doesn't degrade at all the sonic performance of my audio chain.
Last, but not least, our ears behave as perfect natural low pass analog filter, especially for middle-aged people like me.
The analog stage initially acts as single-ended, virtual ground, current to voltage (I-V) converter: in this way the DAC chips (PCM56) can be configured for current output source and be put in parallel, bypassing the inner op-amp and allowing external amplifiers with much better technical features.
The output voltage of a transimpedance amplifier is directly proportional to the current given to its inverting terminal: the gain is established by the feedback resistor and because the amplifier is in an inverting configuration, it has here a value of -1050. Therefore, with a total current of 4mA sourced by four PCM56 in parallel (4 x 1mA), the analog stage will produce an audio signal of 4.2 peak voltage (or 3V RMS @ 1KHz), deliberately higher than the standard of the commercial digital players (2V RMS), as I found this level the best sounding trade-off among gain, dynamic, noise and distortion.
In general, opamp used as IV converter has to slew fast enough to follow the DAC current output. At the same time, the HF energy of the DAC current steps must be absorbed to avoid transient intermodulation (TIM) issues. Here, a damped twin RC passive filter at the opamp inputs, together with a snubber RC network across the feedback resistor, shunt high-frequency energy to ground and prevent the opamp from going into slew-limiting. The compensation network is then optimized to keep the bandwidth wide enough without excessive peaking or oscillation. This pre-filter network is absolutely crucial for the overall "SQ" (sound quality), where the components have been calculated, simulated, listened to, adjusted, evaluated and reviewed again and again until having a truly dynamic and tonal balanced DAC.
The AD828 (from Analog Devices) presents excellent features (bipolar, degenerated NPN differential pair inputs, high speed 130MHz bandwidth, 450V/µs slew rate, 80ns settling time, low 8ohm output impedance, 50mA Class-AB current output) and I've recently discovered to be sonically superior to AD826, providing a wider and more transparent mid-high frequency soundstage. But AD828 is not unity gain stable and can easily oscillate if not correctly compensated at some sacrifice in slew rate: for the stability of the AD828, an additional small capacitor (47pF) was required in parallel with the feedback resistor to safely bring phase margin above 45 degrees.
Also AD826 presents outstanding technical and sounding features: being present in massive quantity (30 chips!) in "The 50th Anniversary Amplifier", it certainly couldn't be missing in this DAC too! Due to its wide band, unity gain stability and unlimited capacitive drive capability, it is perfect in this DAC to be used both as low-pass active filter (a third order Sallen-Key) and as buffer and cable driver, biased in Class-A, in output versus the power amplifier through unbalanced RCA connectors.
The "51st Anniversary DAC" analog stage is soldered in a twin monaural layout for left and right audio channels, with separate regulated power supply paths, an exclusive 25W toroidal transformer with independent windings and extra-large Nichicon 135°C grade capacitors. All low-pass filter capacitors are metallized polypropylene film (by Leclanché, selected pair by me), and the output capacitors are a mix of excellent sounding paper-in-oil and Teflon military Russian components. All resistors are of course 1% grade, selected pair, metal film type.
The BurrBrown PCM56 DAC chips have been massively used in late 80's in several good commercial CD players; audio equipment manufacturers like AIWA, AKAI, DENON, FISHER, HARMAN-KARDON, HITACHI, JVC, KENWOOD, LUXMAN, MITSUBISHI, NEC, ONKYO, PIONEER, SANSUI, SONY, TEAC, TECHNICS, TOSHIBA, YAMAHA have used PCM56 in some of their players.
PCM56 chip performs digital-to-analog conversion using ultra-stable high-precision thin film resistors in a ladder network configuration. Unlike more modern Sigma-Delta converters, which work in 1-bit mode, resistor ladder multibit DAC use hardware to convert the signal and there is neither feedback loop to make constantly error correction, nor a noise shaping circuit to keep the noise away from the audio band. Therefore, resistor ladder multibit DAC provide a better response time and are free of any manipulation in the digital domain, which are probably the reasons why many people consider them sounding more natural and musical than Sigma-Delta.
According to internauts, PCM56P is one of the best sounding multibit DAC chip ever made; when it is used in non-oversampling configuration and put in parallel, for someone it sounds even better than TDA1541 (which unfortunately I never tested) due to a better transparency on female voices.
Paralleling DAC chips in some cases provides specific advantages: a better linearity at low level signals due to a statistical averaging of the conversion errors, a lower noise due to the sum of uncorrelated sources and a reduced distortion due to the need of less amplification in the following stages, thanks to the higher output current available.
To achieve this, four PCM56 chips per channel have been selected by me having the lowest measured distortion (at -80dB) and the most homogeneous output current. In fact the formal tolerance of a PCM56 output current is very high (±30%), which can bring to important differences in levels between the left and the right channel.
All components are coming from mixed lots of premium grade chips (different batches of PCM56-K and PCM56-S selection), in order that bit conversion errors result apparently random and don't skew in the same direction.
In principle this operation should get benefit of running independent processes in parallel, for which the outputs are directly added (because the input signals are correlated), but errors and noise are statistically combined in quadrature, hence, lesser (because they are supposed uncorrelated).
In simple terms, linearity at low level signal should improve because mismatches compensate each other and the errors average; on top of this, to compensate for the total error, only one chip is MSB adjusted at its pins, following the procedure of the datasheet.
Mixing four selected PCM56 chips and adjusting only one of them, the minimum distortion (THD%) that I could get at the DAC output with an input signal of -60dB was as low as 0.38% (or -48dB), which is equivalent to the nominal distortion value of the famous and exclusive Philips TDA1541A/S2, double crown selection, used in the 90s only in really high-end converters.
In total there are eight PCM56, four in parallel for the left and the right channel. I decided to build two power supply boards with separate transformers, bridges and regulators, one for analog circuitry (+/- 5V) and another for digital (+/- 5V) in order to prevent the potentially noisy digital power supply from polluting the clean analogue one.
Two wires for digital and analog grounds are joined in the DAC board site, at each chip level as suggested by the manufacturers: unfortunately I have formed multiple small ground loops due to the parallel configuration of these chips, but this also forms a ground network plane, with some advantages. The analog and digital signals run completely separate on the board and the dual power supply of each chip is further filtered with four damped RCLC networks (damping R = 1 ohm, C = 22uF OS-CON, L = ferrite bead, C = 100nF ceramic for digital, or 220nF polyester for analog).