RME ADI-8 Inside: Architecture, technical specifications and test results

The ADI-8 DS is a reference-quality, 8-Channel A/D and D/A converter. This Tech Info is aimed at technically-minded readers, and covers details which were not mentioned in either the general product description or the User's Guide.

Overview

The ADI-8 PRO internal structure can be divided into 4 distinct functional blocks:

• Analogue In: analogue input stage with gain compensation, A/D-converter
• Analogue Out: D/A converter and analogue output stage with gain compensation
• Digital I/O: ADAT optical and TDIF interfaces
• Clock Control: Controls all timing for all functions, wordclock I/O, PLL

This photo shows the internal structure of an ADI-8, after having opened the case.

As the block diagram clearly shows, the ADI-8 can be routed internally in many ways which, in addition to its function as A/D-D/A converter, allows it to be used as a format converter, distributor or splitter.

• Format converter: 24-bit digital conversion from ADAT optical to TDIF and vice versa.
• Distributor: Sends digital input to all outputs simultaneously, e.g. TDIF in to TDIF out and 2 x ADAT.
• Splitter: Sends analogue input to all digital outputs simultaneously, i.e. for A/D conversion to 2 x TDIF and 2 x ADAT (optical)

Analog to Digital

Physical access to the analogue inputs is either via 1/4" stereo jack socket or 25-pin D-type connector conforming to the Tascam DAxx analogue I/O pinout. These recorders are very common, so a suitable multicore cable (e.g. for connecting directly to mixing consoles or to XLR) should be relatively easy to find. The entire input stage - from the sockets to the converter chip's internal input - is fully symmetrical. This is a special servo-balanced input circuit: if the signal at the input is unbalanced, input gain is automatically adjusted by +6 dB. Otherwise input sensitivity for an unbalanced signal would have been lower than for a balanced one.

Any A/D converter is only as good as the gain compensation at the input. The ADI-8 has 3 gain levels, making it compatible with a wide range of studio equipment - loss is avoided before the signal gets to the A/D converter. Input gain is switchable between Lo Gain, +4 dBu and -10 dBV, and employs a special damping technique: using latest hi-grade electronic switches (ADG 451), the gain switch could be placed in the best possible position within the signal path. This approach makes additional amplifier circuitry or long cables to mechanical components unnecessary (and obsolete.) The D/A converters get the best quality signal possible, so they can reach maximum SNR (signal-to-noise ratio) with minimum distortion.

Fig. 1 shows the noise floor level (as SNR) at all 8 inputs in Lo Gain position, measured RMS unweighted. A-weighted, the results are about 3 dB better. In the -10 dBV position, the signal is amplified considerably, so input sensitivity is naturally higher. Nevertheless, SNR still attains 107 dB RMS unweighted.

Low noise JRC4580 op-amps are used throughout. This type can be found in countless appliances - it is an unbeatable combination of low noise, low output impedance and low price. However, simply sticking 4580s into your devices will not guarantee you the best possible technical spec - the surrounding circuitry has to be up to scratch as well.

We decided on AKM5392 as the A/D converters (ADI-8 PRO). These high-end components are relatively expensive, but on the other hand they are almost unbeatable quality. The combination of our special fully-symmetrical input stage with these converters is the main reason for the exceptional 112 dB SNR RMS unweighted (115 dBA), as well as extremely low distortion (THD), below -110 dB (0.0003 %.) Remember that these are not just phantom values copied from the component manufacturer's' spec-sheets - they are the results of real tests on a complete product in a standard environment.

Fig. 2 shows the Dynamic Range when a 1 kHz sine wave (from a Neutrik A2D) is fed to the analogue input. The resulting digital output is -60 dBFS. It also shows zero distortion and a very low noise floor.

The reason for this test was to detect noise modulation (which is only triggered by an input signal and does not appear otherwise). Noise modulation was undetectable here - the noise floor stayed at exactly the same level (about -150 dBFS.)

Things started getting really interesting when we checked the ADI-8 PRO for distortion caused by the A/D converters. Our low-jitter design, the fully symmetrical input stage and the excellent AKM converters caused an unusual problem: Any distortion we could measure came from whatever signal generator we were using for the test - we detected no distortion attributable to the ADI-8 PRO's input. Did we go out and buy another expensive audio analyzer just for this test? No, we did something that many people in this business would normally avoid like the plague: we used the ADI-8 PRO's own analogue output as the signal source. Normally, this would completely ruin any chance of good results for the A/D converter - but the ADI-8 PRO is not a normal machine ... see below for more about this incredibly high-quality D/A conversion.

A very pure sine-wave, digitally generated by the audio analysis program HpW Works, was converted to analogue by the ADI-8 PRO's D/A, fed to the ADI-8 PRO's input, and then digitally analyzed via FFT. The many improvements we made to details certainly paid off - The HpW Works analyzer showed no less than 111 dB THD! We popped a few champagne corks then ...

Fig. 3 shows distortion measurements with (THD+N) and without (THD) the noise component. A sine-wave just below maximum level with harmonics well below -110 dBFS clearly shows that the ADI-8's claim to reference-quality is justified.

A THD+N of well under 0.00074 % puts the ADI-8 firmly into the $5.000-and-over league.

Note that the quality of the A/D conversion on its own would be about 3 dB better than this combined A/D-D/A test. The THD+N should therefore be about -105 dB, SNR should be - assuming a perfect signal generator - much lower (about 5 dB), because the noise component from the D/A converter is more evident here. Take another look at Fig. 2, where a very low-noise signal was used.

Jitter

Experts will have recognized it immediately: the small bell-shaped curve at the base of the 1 kHz sine is caused by jitter. Usually nothing special - but in this case it is doubly so!

First of all, this jitter effect would normally be invisible. Compared to the FFT analysis functions in HpW Works (the program RME uses for most of its digital tests), those you find in hardware audio analyzers - even those with luxury-car price tags - are relatively poor. Instead of 65,536 points, the resolution is restricted to 16,384. Also, there is seldom enough windowing, usually essential for FFT. Remote damping is insufficient, and so the base of the sine-wave is so wide that even much higher levels of jitter than in fig. 3 still wouldn't register.

Secondly, the jitter you see in fig. 3 is very low. This is not interface jitter from the AES/EBU or ADAT interface, but so-called 'Sample Clock Jitter', which chiefly depends on circuit layout, the jitter suppression in the A/D converter and the device's master clock. Completely jitter-free systems are physically impossible, but bad layout and/or a poorly conceived clock can ruin the excellent results that modern converters are able to deliver in the right environment. As you can see in fig. 2, jitter products lie below -135 dBFS. Because this value depends on signal amplitude (i.e. it will disappear under the noise floor if the signal is low enough), we can safely call jitter at such low levels irrelevant.

Testing sample clock jitter seems to be fashionable these days, because advertising can capitalize on the (apparently) sensational values of a few picoseconds. Fig. 4 shows the results of such a test (10 kHz sine FFT.)

We are not sure whether the slight distortion at 20 kHz came from the ADI-8 PRO's D/A converter or from its A/D. However, at -107 dBFS this is not a particularly interesting question...

With interface jitter, we are usually talking in terms of nanoseconds. Sample clock jitter is much shorter still - the ADI-8 attains spectacular values of under 20 ps (picoseconds, 10 to the power of minus 12.) However, jitter naturally depends on the clock mode being used. Fig. 4 illustrates a worst-case situation, the ADI-8 A/D set to INP (slave clock mode), so it is being clocked by an external device (in this case a DIGI96/8.)

As would be expected, results of jitter are lower in the (for A/D-D/A converters) preferred INT/INT configuration. Fig. 5 shows that modulation products are well below -130 dBFS, and therefore absolutely meaningless.

Looking at these two results, you might be tempted to run only INT/INT in future. But to put things in perspective: individual peaks at around -125 dBFS (see frequencies below 2 kHz) are completely insignificant when the signal itself is 120 dB higher. The dimensions become clearer when you look at the values for THD+N and DR (see fig. 4 and fig. 5) - they are almost identical.

In principle you could get a cleaner result (even damping the peaks below 2 kHz) by using a particularly slow i.e. highly filtered and/or quarz-based PLL. We have very good reason for not even trying these out. Our bitclock PLL requires a minimum rate to track ADAT XT Varipitch consistently, i.e. without any dropouts. Nevertheless, it still has excellent jitter-damping, beating wordclock-based PLLs by far. Quarz-based or highly filtered PLLs react adversely to rapid changes in sample frequency, causing dropouts (clicks/mute) - several samples will be read falsely before the PLL adjusts to the new sample frequency. This is much more of a problem for the 8-Channel ADAT format than for the 2-Channel S/PDIF or AES/EBU formats.

The frequency response of ADI-8 PRO's analogue input is flawless (see below.) However, thanks to multiple oversampling, only very poorly designed pieces of equipment would exhibit any significant deviation these days. In other words, frequency response curves (as seen in audio journals) of any modern professional equipment - not only ours - are and should be absolutely flat.

No Dither?

No Dither! Dither is not a simple exercise in a device which has been realized using only FPGA. Besides, in certain cases it can be dispensed with altogether. If you would like to know more about this view (which some might consider provocative), please read the Tech Info Remarks about the Need for Dither.

Digital to Analog

In the first ADI-8 PRO version we used the Analog Devices AD1855 chip as D/A converter, because this was the first DAC to achieve 110 dB dynamic range without resorting to muting to improve S/N ratio. At the same time, the AD1855 is relatively cheap, which is especially important for multichannel converters such as the ADI-8 PRO, with its 4 converter chips.

Since we brought out the ADI-8 PRO revision 2, we have been using the brand new AD1852, a chip which drives testers up the wall (because they are such good quality - see above.) RME is the first manufacturer ever to market products using this DAC. Unlike the 1855, the 1852 is even less susceptible to jitter and high frequency clock interference, and the distortion factor has improved by more than 10 dB. The former ensures 3 to 6 dB higher SNR in various clock modes, while the latter allows the ADI-8 to carry out its unique 'selftest' as described above. Because the 1852 is pin-compatible with the 1855, we started using the newer chips as soon as they became available.

The D/A converter sends its output to a symmetrical filter to remove aliasing frequencies. The signal is then fed to an intermediate stage to adjust the gain (using the same IC switches as in the input stage) before reaching the servo-balanced output stage. The output stage has been calibrated (a trimmer) for full symmetry before leaving the factory. The output stage also automatically compensates non-symmetrical output (by 6 dB like in the input stage). The 4580 op-amps are not only very low-noise components, but are also excellent line drivers, so a low-impedance, clean output signal is available at the 25-pin socket as well as the stereo jacks. Both outputs can be used at the same time.

You've seen this before? Right, Fig. 6 is actually the same as Fig. 3, and shows D/A as well as A/D conversion in the ADI-8. We don't have an exact figure for just how low the D/A distortion is, but we are already perfectly satisfied with these results, although they include additional errors from the A/D conversion:

0.00028 % THD, 0.00074 % THD+N - what more do you want?

Admittedly, we used a particularly good D/A channel during this test. Analog Devices specify for their AD1852 a THD of 'only' < -104 dB, but in reality values for individual components vary between 105 and 112 dB. Therefore we declared a value of < -105 in our ADI-8 PRO spec. As far as noise from the ADI-8 D/A is concerned, results from a Neutrik A2D ranged from 106 to 110 dB RMS unweighted (depending on the clock configuration and source), or 108 to 112 dBA.

Like the inputs, the ADI-8 PRO revision 2 analogue outputs also have a 3-stage gain switch. The same rule also applies here: Intelligently designed gain compensation is the only practical way to ensure maximum performance. The ADI-8 PRO output stage, with its electronic switching and optimized low-noise output circuitry, reaches the maximum SNR of 110 dB RMS unweighted at +4 dBu and Hi Gain. In the -10 dBV position, SNR is still 104 dB RMS unweighted. The difference is no reason for concern here - 104 dB still means that the ADI-8 PRO is reference-quality.

You would be hard pressed to find a -10 dBV based device which can deliver more than 96 dB. And converters without gain compensation have to accept the loss (from maximum level to -10 dBV) as additional noise. That would have meant - for an ADI-8 without gain compensation - a SNR of only 95 dB.

Crosstalk

When audio signals are as clean as this, crosstalk should also be extremely low. Crosstalk can be measured by feeding one channel with a 1 kHz sine at the highest level possible, and then measuring the RMS level in a neighboring channel. The 1 kHz crosstalk signal is isolated using a narrow bandpass filter, removing noise before the level is measured (because crosstalk can be well below the noise floor in a well-designed system). For the same reason, the input of the channel being measured is also grounded.

We carried out conventional measurements of crosstalk in the ADI-8 PRO's D/A using a Neutrik A2D. Internally, the 8 channels are actually 4 stereo channels, and crosstalk damping between 2 stereo channels (i. e. within the same DAC) amounted to 110 dB. This value corresponds to maximum attainable SNR, and is therefore considered to be very good. However, we were astonished by the results of testing for crosstalk across stereo pairs e.g. between channels 2 and 3: No crosstalk whatsoever could be detected here!

After that we repeated the tests with FFT in the digital domain, and arrived at the same results. The FFT is able to separate the noise floor into its component frequencies at about -140 dBFS, and no 1 kHz peak was found. Crosstalk damping across stereo pairs (i.e. different DACs) was at least 140 dB, confirming just how good the circuit and printed board layout is.

Fig. 7 shows the results of the A/D crosstalk test. Values attained here also range from 'undetectable' to -132 dBFS.

Frequency Response

Fig. 8 shows the frequency responses of A/D and D/A conversion. In the past, anti-aliasing filters used to cause ripple of up to 0.5 dB in the highest frequencies. Nowadays, A/D and D/A converters use oversampling and sophisticated digital filters to ensure phenomenal linearity.

Note the high vertical resolution in this test, 1 dB per grid line.

TDIF

The ADI-8's TDIF connections support 24-bit audio and emphasis (since revision 2), and run without any word clock problems of any kind. You are free to use word clock of course, but in most cases this is not necessary. We have also dispensed with jumpers and other configuration aids for different operating modes, as they are completely unnecessary in a well-designed TDIF interface. There is only one restriction to the above rule (caused by the necessity to support the DA-88/38): If the recorder is the slave, the ADI-8 PRO's word clock output MUST be connected to the recorder's word clock input. The other way around no word clock connection is needed - despite what some manufacturer's (and therefore many users) may claim.

ADAT optical

The ADAT optical connections in the ADI-8 PRO support 24-bit audio, and run without any word clock problems of any kind. Of course you are free to use word clock - but in most cases this is not necessary. The ADI-8 PRO always uses RME's bitclock PLL (already in its second generation, with improved HF noise suppression), whether it is synchronized via Optical In or via word clock. Thanks to the high resolution, this remarkable circuit is capable of following the entire range of an ADAT recorder's Varipitch without losing a single sample. Compare this with the much coarser wordclock PLL, where any rapid changes in sample frequency can cause several false reads, before the next impulse finally arrives, letting the PLL track the current sample frequency. This behavior leads to dropouts and clicks. So as long as you use an ADI-8 PRO, anything goes. With other brands it's a case of 'suck it and see'...

SyncAlign

SyncAlign embraces a whole bunch of techniques to guarantee absolutely perfect timing and no channel problems. SyncAlign was first realized in the DIGI96 series, and of course we have implemented it in the ADI-8 PRO as well. The ADI-8 will NEVER mix up channels, no matter which clock mode is being used. SyncAlign also regulates synchronization between channels (the screenshot below illustrates the sample-sync capability - it shows individual samples of a single 1 kHz sine-wave sent to all channels.) Even when several ADI-8 PROs are synchronized via word clock, the converted signal always appears absolutely in sync across all digital outputs.

Clock Control

The various functions in the ADI-8 are all handled by an FPGA (field programmable gate array) control unit. If you don't know what an FPGA is, imagine it to be a kind of fixed-program microcontroller. This chip controls the sample frequency of both A/D and D/A, their switching, evaluation and display of clock synchronism (SyncCheck), evaluation and display of analogue levels, and much more. It is also responsible for managing audio data, e.g. format conversion, splitting and mixing, as well as audio buffering (see the section 'Delay' below.)

We are especially pleased with the ADI-8 PRO's unique (so far) user interface. The buzzword here is 'ICC' (Intelligent Clock Control), which describes the ADI-8 PRO's intelligent user interface and internal control for its complex clock-options. For the user this means: Switch on, configure within seconds, recognize and correct errors immediately (LEDs blink if lock is absent or there are any synchronization problems) - ready!

This is the way it should be - but without an ADI-8 PRO, it usually isn't.

Delay

Any digital device causes a delay between input and output. As the ADI-8 PRO does not contain any DSPs, there are only two relevant delays:

Digital:

Both ADAT inputs as well as the Copy Mode use a sample buffer. All incoming ADAT data is delayed by 1 sample. Thanks to RME's special 'Virtual Sample Buffer' technology, the TDIF input has no delay at all. In Copy Mode, incoming TDIF data is delayed by 1 sample before being sent to ADAT/TDIF, incoming ADAT data is delayed by a total of 2 samples.

Analog:

Delay from A/D and D/A conversion depends on the chip being used, the ADI-8 PRO itself doesn't add any. Delay within converters is caused by the oversampling and subsequent digital filtering. AKM5392: 39 samples, AD1852: 43 samples.

The Power Supply

When designing a power supply for a combined A/D-D/A device, you have to bear a lot of criteria in mind. While looking for the best solution, we soon decided on a switch-mode PSU. The first advantage that springs to mind is that they automatically accept 100 to 240V (the international range of mains power). This is very important from the manufacturer's point of view - we only need to build a single model for the entire international market.

Another, equally important topic is power consumption. The ADI-8 PRO is chock-a-block with components, so it draws quite a lot of current. A conventional PSU would not only take up more room, but would also generate more heat and quite a lot of stray magnetism. These drawbacks do not apply to switch-mode PSUs, which are very energy-efficient. The type we built into the ADI-8 PRO has even more to offer: it is absolutely short-circuit-proof, has an integrated line-filter, is fully regulated against voltage fluctuations, and suppresses mains interference. These are all aspects which guarantee that the ADI-8 will work reliably at all times, even under adverse conditions.

Using switch-mode PSUs, there is always the (justifiable) fear that switching transients could cause pickup and other stray effects in the analogue circuitry. Besides, the voltages this type of PSU delivers are generally not as clean as conventional transformer power supplies. These are all legitimate fears - but they don't apply to the ADI-8. For proof of this, see the A/D and D/A test results above.

Remarks about the Tests

RME carries out all tests using professional measuring instruments, adhering to standardized metrological guidelines. We consider it important to arrive at real results under real test conditions. Some of the competition prefers to copy their data straight from chip-manufacturer's' spec-sheets. Such fantasy-results are not only deceitful, but are also impossible to reach - for two reasons: Firstly, they don't take into account all the other active electronic components in the finished product. Secondly, it is common knowledge that quite a few chip specs themselves are 'ultimate goal' values, only reached after several revisions - if at all.

Double-checking these results by plowing through magazines looking for RME product tests is unnecessary - these are real tests and real results.

We use the Neutrik A2D and its accompanying software AS04 (www.neutrik.com) for our analog and digital measurements. This yields (among other things) RMS unweighted or weighted results according to defined industrial standards, for levels, distortion and noise, as well as sweeps for nearly all parameters (see blurb about frequency response). The A2D is a must for hardware developers thanks to its simple operation, high-precision results, and its connectivity.

For purely digital measurements, we use a reference-level Windows program called HpW Works. This software has a very high resolution and delivers unbeatably accurate results. Apart from the FFT analyses mentioned above, it can also be used to generate highly accurate digital signals.

We used our own DIGI96 series (of course!) to interface with the PCs, so we could be absolutely sure that all data is transferred one-to-one with no errors. These also serve as SPDIF to ADAT converters (and vice versa), without which many of the tests would have been impossible - even some of the most expensive audio test systems do not support ADAT, but are purely SPDIF and AES/EBU.

Glossary

16-Bit: Equals a resolution of 65536 level steps or theoretical 96 dB dynamic range.

20 Bit: Equals a resolution of 1048580 level steps or theoretical 120 dB dynamic range.

24-Bit: 16,777,200 level steps or 144 dB theoretical dynamic range.

ADAT optical: TOSLINK interface using the Alesis protocol. Allows maximum 8 channels at 24 bit resolution and 48 kHz sample rate.

dB(A): RMS value with A-weighting filter. Provides a response being less sensitive at very low and very high frequencies.

dBFS: dB Full Scale. Logarithmic level value relative to (digital) maximum level.

DR: Dynamic Range. Ratio of maximum level to noise floor, with a trigger signal at -60 dB.

FFT Analysis: Fast Fourier Transformation. Spectral display of any signal.

Jitter: Time fluctuations in signals.

RMS unweighted: Effective value (RMS) using a bandpass filter (22 Hz to 22 kHz.)

Sample frequency: Number of samples per second.

Servo balanced: Circuit able to operate balanced or unbalanced, automatically correcting gain.

SNR: Signal to Noise Ratio. Ratio of maximum level to noise floor.

THD: Total Harmonic Distortion. Ratio of harmonics 2 - 10 (summed) to the fundamental.

THD+N: RMS measurement of all harmonics plus noise relative to the sum of all signals together.

Copyright © Matthias Carstens.

All entries in this Tech Infopaper have been thoroughly checked, however no guarantee for correctness can be given. RME cannot be held responsible for any misleading or incorrect information provided throughout this document. Lending or copying any part or the complete document or its contents is only possible with the written permission from RME.