Dr. David Rich on DACs, including the one used in the DVD-S1500.

Dr. David Rich on DACs, including the one used in the DVD-S1500: The CS4382 DAC delivers better than average behavior. To make life easier for our readers I am converting specification from the data sheet into effective number of bits (ENOBS). From the data sheet we find the CS4382 has typical 19-bit performance for noise, and with worst case 18-bit performance. When a manufacture states that a parameter is "worst case," this means that another manufacturer using the part in his product can test that part and expect it to meet the specification.

Unfortunately, a specification marked "typical" on a specification sheet can be a problem, because many manufacturers do not say what "typical" means. However, any decent designer designs only with worst-case numbers in mind if the parts performance for what he is designing can affect the stated specifications of the final system. For example, the motor in a DVD player had better be able to spin fast enough so the player can read the DVD. A "typical" spin speed in the motor's specification world not cut it. In other words, the designer would want the motor manufacturer to supply a worst-case motor spin speed to make sure that the player can always meet minimum standards.

With some DA converter specification listings, only the "typical" specs are given. If the final-product manufacturer uses those chips and has done no assembly line testing to insure that DVD players delivered to consumers meet some minimum specification, many of those players may not do so. This is the case, unless the manufacture indicates the specification is guaranteed, as is the case under federal law for amplifier power output under a specified load, frequency range and THD. To my knowledge, the only hi-fi company that currently states that the all the specifications they issue are guaranteed minimums (or maximums) is Accuphase.

The noise figures I have just given in ENOBS are measured by Cirrus Logic with an "A-weighted" filter. That kind of weighting is applied by placing a filter before the noise meter. The filter is said to allow the measurement to better reflect how the ear perceives noise level changes, meaning less sensitivity at the high and low end. Weighting also makes DAC noise performance look better since some noise has been filtered away. However, it really has no application here, since a properly designed DAC should have a flat noise floor, although cheap DACs can have an increase in noise at the low end. This is a sign of a design compromise that we do not want hidden under an A-weighted rug.

Even when no weighting is used, the DAC in this player has a little more than 17-bit resolution. Distortion at -20 dB input level is a little better than 18 bits typical. At full scale it drops back to 16.5 bits. Worst-case, full-scale distortion is slightly below 16 bits. These distortion numbers are for 1 kHz. The numbers usually get worse at higher frequencies, but the data sheet does not list distortion results at higher frequencies. The decline of the ENOBs with increasing input signal level frequency at full scale is one of the key benchmark tests for evaluating the usefulness of a general purpose DAC in a particular application space. For an application specific chip like the CS4382, the information sometimes does not make it to the data sheet.

That said, Cirrus supplies almost all other dynamic specification about this chip, whereas data sheets for chips that do not perform as well, but are likely less expensive, may contain no worst-case numbers, and numbers without A weighting will not be listed by manufacturers. Providing more detail on the data sheet shows the company has confidence in the design's ability to deliver these numbers when the devices are mass produced.

How many bits of noise level headroom do we need? Well that depends on how quiet your room is and the maximum loudness you will tolerate in that room. (This is the all-digital-ones level of the CD--the largest signal level the DAC can reproduce.)

Of course, other components in one's audio system must also have noise specs as good the DA converters. Most home electronics will not hack it, since a maximum signal to noise ratio of 110 dB is the equivalent of 18 bits. And of course the recordings must also have been created with very low microphone noise. In addition, they must have low noise in the analog signal path that follows the microphone (analog level adjustment and equalization may sometimes be used in the production of a modern CD), have low noise in the studio or concert hall, and also have sufficiently low A/D converter noise.

Data in conference and journals have presented information that points to 18-bit equivalent signal-to-noise levels as the required minimum for a professional studio (designed for very low background noise, which may only be achieved with special construction techniques and materials). I do not recall the maximum signal level (all ones) was used for the tests but I assume it was at least movie theater level loud.

With respect to distortion, again we must consider the rest of the equipment in the system. Almost all of you have seen THD vs. level graphs and will recall they rise as the level gets higher. At higher signal levels, more nonlinear effects of the electronics are uncovered. This is true with the DAC as well as analog components. In most cases the power amp will dominate a systems-distortion level at maximum signal level - the point where the power amp is about to clip, which in a digital system should be set to correspond to the all ones digital representation of the loudest signal level on a CD. Only the very best power amps could match this Cirrus converter's distortion level at its worst-case distortion specification.

Before we get too excited, please recall the DAC in a DVD player is only in use in SACD or DVD-A modes when the analog pass-through of your AV receiver is active. In this mode, all the good things your AV receiver can do (advanced digital bass management, multi-band EQ, multi-channel synthesis etc) are bypassed. When playing normal CDs (via the digital hookups to the receiver and not the analog outputs) and DVDs, it is the DA converters in the AV receiver that count and not the one in the player.

This Cirrus chip has a balanced output, and this is found only on the better converters and requires more analog electronics. The digital filter preceding the DAC is a complex design providing a digital frequency response of +/-0.01 dB and a 90-dB stop band for digital signals at a 44.1kHz sample rate. Although the +/-0.01 dB spec noted in the Cirrus info sheet may look silly, it is an important indicator of FIR filter tap length, and it correlates with the very important stop band attenuation. The best chips are +/-0.002 dB with more than 100 dB of stop-band rejection. The CS4382 chip has a true DSD inputs for the SACD disks. However, I cannot tell whether it passes it through to the analog output directly or turns it first into PCM, in which case any advantage of SACD signals having no digital signal processing is rendered moot. The data sheet is unclear--in one section on the frequency response of the chip I find the heading "Combined digital and on chip analog filter response -DSD mode."

The chip also has a slow-roll-off mode that trades stop-band attenuation for improved group delay flatness in the passband (improved by a factor of 3). In the 44-kHz mode the slow rolloff starts slightly in-band, at 18.3 kHz instead of 20 kHz. Both fast and slow rolloff modes bring the signal level down 3 dB at 21.9 kHz. More significant is the change in the rejection of the 20 kHz first-folding tone that results in the reconstruction process of the sampled signal (24 kHz for a CD). This is down only 20 dB, which is 10 times the value of the 40 dB in the fast mode. (Remember, decibels are in logs; hence a doubling in decibels is a 10x increase.) Maybe a teenage kid can hear it. The slow mode will make the ringing of the filter to an impulse look better in the time domain but at the cost of a potential audible effect, at least for teens. In addition I note that no scientific study has shown the ear is not sensitive to group-delay flatness.

In the 96-kHz mode, the rolloff moves from 42 kHz to 28 kHz. The -3db point is constant at 48.9 kHz. In the 96-kHz mode the first fold tone for 20 kHz is 78 kHz and this is well rejected with both fast and slow filters (and your ears). In the 96-kHz mode, group delay flatness is 14 times better than CD with the filter in the same mode. As was the case for CD, the slow mode makes things three times better. However, now we are starting with very small variations in group-delay flatness as a result of the reduced requirements on the filter to have an extremely steep transition band when the signal is sampled at 96 kHz.

The effect of the shape of the impulse response of the filter in the time domain that results from moving from 44 kHz to 96 kHz sampling is easier is to appreciate directly in comparison to looking at group delay curves vs. frequency. Some people I know who are experts in sampled data systems (but not in audio) say the reduction in the ringing before and after the impulse might have some effect on the reconstruction of signal in the time domain. Audio Engineering Society conference papers have been presented giving more details on this, but I have not seen them make it to the society's Journal, which critiques the materials in greater detail prior to acceptance.

The fast and slow filter responses will change the shape of the impulse. I have no recommendations on which shape would be preferable. 96-kHz sampling is such overkill that small details of filter rolloff become inconsequential. From an engineering point of view, that is a good thing. The difference between reconstructing a signal at 96 kHz and reconstructing it at 44 kHz is clear for all to see on a scope. Whether the difference can be heard is still an open question, but given a choice I will go for the thing that measures well, if it does not cost me any more. In the case of the hardware, the cost impact is small. Software costs are still a major issue.

The bottom line with this DAC is that you lose only a bit in comparison to the best. However, you have at least an extra bit over the lower-cost universal DVD players as well as the AV receivers I am currently testing. The players I am currently testing are, under worst case specs, just 16-bit engines. However, that is all you need for CD playback, and that type of performance in mid-line products would have been impossible a few years ago.

Conclusion: at $450 this unit is a good deal without question, assuming the analog stage is not messed up.

--DAR