Considerations and Personal PreferencesNo PCM but The Sound of Oversampling / Upsampling and Noise Shaping
THE COMPACT DISC Philips developed the basis for the current Compact Disc and Sony joined during the last years when Matsushita decided not to work with the Philips engineers. The Philips engineers developped the 120 mm (4.7 inches) disc which holds the original signal and the various corrections which are necessary to read the stored signal correctly. This certainly was no mean task. It is a clever storage medium.
Since Philips initially developed a converter with 14 bit quantisation that worked with the chosen sampling frequency of 44.1 kHz., Sony insisted that the quantisation should at least have 16 bit because severe criticism was expected, especially from audiophiles and serious music lovers, who by the end of the nineteen seventies, possessed a phenomenal sound reproduction of analogue recordings on records and tapes:
1.
a high resolution of every frequency in the audio band,
2.
an extended (practically unlimited) frequency band for upper harmonics, and
3.
a very good transient response,
4.
good dynamics,
5.
an extremely low noise level, and
6.
minimal distortion.
NOISE SHAPING A 14 bit quantisation results in 16.384 pulses per second. With 16 bits there are 65.536 pulses which is quite an improvement. Despite the advantages of digital recording, for many the format of the Compact Disc of 44.1 kHz. and 16 bit was a serious disappointment as far as resolution and frequency band are concerned.
In order to be able to comply with the 16 bit criterium Philips applied the technique of four times oversampling and incorporated a noise shaping filter to its 14 bit converter. The addition of the noise shaping filter resulted in an outcome which was close to the 16 bit signal to noise ratio of 96 dB (each bit represents 6 dB0.
The noise shaping however was noticeable in these first Philips players and especially the high frequencies did not have a clean defined signal but appeared to sound somewhat diffuse. Important in this matter however is the order of the noishaping filter and the audible effect. Lateron the filtering was improved. But there always remained a difference between a datastream with noise shaping and a datastream without noiseshaping.
LOW BIT The advent of high speed processors made the creation of so called low bit converters possible. Initially they were meant for portable digital equipment: CD-players and DAT-recorders for outdoor use by joggers and people travelling. Low bit converters are very practical and less costly to produce than the multibit converters. Low bit converters do not need a precise adjustment of the bits, especially the most significant bit as is the case with multibit converters. Multibit converters originally kept their place in stationary quality CD-players and DAT-recorders in the studios and homes.
PULSE DENSITY There are two version of low bit conversion techniques. Matsu######a's conversion is based on the so called Pulse Width Modulation. The total music signal of 16 bits is being contained in a quantity of a few bits (actually 3.5 bits) which do not ask for precise adjustment of any bit.
Philips developed their own conversion based on Pulse Density Modulation. There the complete signal is contained in just 1 bit. Both manufacturers began to incorporate these cheaper converters also in machines for home audio.
These conversion techniques showed an improved wave form at very low recording levels where multibit converters show a ladder instead of a sine wave.
However the improved sine wave generated by a single bit converter (be it based on Pulse Width Modulation or Pulse Density Modulation) is not a thin fine line as normally should be the case, but the wave is a thick, somewhat wooly sine wave which indicates that some manipulation of the data and a high oversampling rate have been applied.
FILTER When choosing a sampling frequency of 44.1 kHz., only the frequency band of 0 to 22.05 kHz. can be used because this band is mirrored by the band above the 22.05 kHz., namely the band between the frequencies from 22.05 kHz. to 44.1 kHz.
This means that each and every frequency under 22.05 kHz. has an alias in the frequency band above 22.05 kHz. Hence the term aliasing.
It is important that the band of 22.050 kHz. to 44.1 kHz. does not interfere with the actual audio band. In order to make this interference impossible, a steep filter (brickwall filter) must be used. No aliasing may occur.
By applying an oversampling rate of four times (or a multiple of four times), the brickwall filter is not necessary to act at at 22.050 Hz., the frequency at which the filtering normally should take place. Because of the oversampling, a less steep filter can be applied, say 6 dB at 22.5 kHz.. Now not only a wider frequency band is suggested, but also the phase of the signal has improved. The fine detail which is a trademark of a real wide frequency band is however missing.
DIFFERENCES There are technical differences between the precise multibit converter and the low bit converter. These differences are audible too. Listening to a recording of a symphony orchestra or an ensemble converted by a multibit decoder the space between the instrumentalists is apparent. One can, as it were, see the stage floor. Denon, Accuphase and other high-end brands like Mark Levinson, Theta, PS Audio, applied precise multibit conversion. Multibit converters have to be adjusted very precise in order to keep the harmonic distorsion very low and to achieve a linear frequency response. Both are related to the accurate adjustment of the bits. The measure of oversampling (4x, 16x, 32x, because designers started applying oversampling in multibit converters first and at a lower rate) determines the accuracy of the signal, especially the accuracy of the high frequencies. Personally I find 8 times oversampling an optimum for the ear. In this way the sampling frequency is transposed to 352.800 Hz. An oversampling of 16 is to my ears the limit. Higher rates of 32 (Wadia), 64 (Krell) and of course 256 times make an audible difference and deliver a less "natural" quality.
MULTIBIT AND MULTIBIT Listening to a musical signal that was converted by a precise multibit decoder "shows" that the instrumentalists are seated in front of each other, next to each other and behind each other, which means they are seated at random so to say as on the stage. When using a multibit converter which has noise shaping (as with players of Philips and Marantz) it seems as if one looks over the heads of the instrumentalists. The space in between the instrumentalists is less evident. I personally noticed many times that a low bit converter places the musicians as if seated neatly in half a circle which is not the reality.
On top of that the attaque in the musical passages with strong dynamics and complex structures, do sound a grade weaker, they spread out, so to speak. Through a high rate of oversampling, combined with the noise shaping, the highest frequencies sound friendlier but are less chiseled and less "clean". The advantage of this concept however is that the nasty ringing of a steep filter is being avoided and a better transient response is suggested.
FURTHER DEVELOPMENTS When in 1994 Pioneer Electronics introduced their Wide Band DAT-Recorder with a sampling frequency of 96 kHz. and 32 bit quantisation, Philips technicians were immediately interested. Right after the press conference the machine was sent to Eindhoven at once. The engineers' interest was brought about by the fact that a higher resolution was achieved than normally was the case in a DAT recorder. But there was another peculiarity. Pioneer had used 2 Philips bitstream (1-bit) converters based on 16 bit quantisation and a sampling frequency of 48 kHz. They worked together and thus a wide band was being achieved. The application of a very expensive 32 bit converter (as was already used in the Mitsubishi Digital Tape Recorder in the nineteen eighties), was being avoided. Using bitstream converters in this DAT-recorder opened new and inexpensive possibilities. Nevertheless, the fact that there was noise shaping could not be denied when listening to the Japanese train passing by and the birds chirping in the trees.
SUPER AUDIO CD The design of high speed processors made it possible to develop the low bit conversion technique could be developped further. The result was Direct Stream Digital (DSD) which is the basis of the Super Audio CD.
In Positive Feedback (Vol. 8, No. 2) designer Ed Meitner is being interviewed by Mike Pappas. Meitner was involved in the implementation of the Super Audio CD. He accuses those technicians of the early nineteen eighties who did not listen carefully enough to the new CD format because they were completely caught by the novelty of the digital technique. Since then they have become so familiar with the "simple" format of 44.1 kHz. and 16 bit that they now do not see the importance of the high resolution which DSD brings about.
Furthermore Meitner states the well known adagio that "less" is always better than "more". In this respect he talks about circuits and buffer amps in players that should have a discrete lay-out as a result of the high energies which are generated by the vast datastream. Integrated circuits and certain opamps cannot deal with these high energies, he says.
However if one sees the complexity and the manipulation of the signal that takes place in the converter of the Super Audio CD one must conclude that Meitner contradicts himself. It is not less but more.
HIGH END? In defending the Direct Stream Digital of Super Audio CD Meitner says in the telephone conversation with Pappas that some high-end manufacturers use low bit converters in their very costly DA-Converters. So why should not they use Direct Stream Digital?
The use of low bit converters in high-end players is of course an idiocy. Many times we have witnessed demonstrations with DACs from Audio Research and Threshold at the time. When I asked the demonstrator if it was possible to connect a multi bit DAC, in most cases they were able to produce such a component. For all listeners it was evident that the five times more expensive low bitters could not match the quality of the multi bit DAC.
So the use of low bit converters in high end machines and seperates is not an argument for anyone to abolish PCM and multibit conversion and switch to DSD and low bit conversion. On the contrary! Furthermore the use of 1-bit conversion in expensive components (Threshold, Audio Research) is not proof that it provides a better outcome of the treatment of the data compared to what a multibit converter does. Just listen to the Accuphase technology.
Personally I swear to precise multibit converters as applied by Denon and high-end manufacturers like Accuphase, Theta, Enlightend Audio, etc. And personally I would have given my preference to a Super Audio CD if the technique would have been based on Puls Code Modulation but with a high sampling frequency and a high level of quantisation with precise multi bit conversion. As said earlier, Mitsubishi in the nineteen eighties, already had a digital reel to reel recorder and AD and DA converters with a sampling frequency of 96 kHz. and 32 bit quantisation. To-day even higher frequencies and bit rates would be feasable.
All this does not mean that one would not see the inginuity of the DSD as everybody marvelled at the technique of the old CD format.
GIGANTIC Through oversampling the conversion frequency of the Super Audio CD is 2.8224 MHz. as stated by the datasheets. Such a high frequency gives a better resolution if compared to the current PCM formats. Because of the fact that SACD uses DSD and not PCM, the frequency band is not half of the 2.8224 MHz. but extends to just over 100 kHz. (110.25 kHz. I suspect, no accurate frequency is mentioned by Sony and Philips). This suggests that the oversampling rate is 256 times.
The designers of DSD bitstream state that it has two more advantages: a low harmonic distorsion and a perfect linearity.
But also regarding these two aspects, the precise multibit converters do show an extremely high performance. The ingenious technology of Accuphase regarding the current CD format for example does not only show this in measurement but let you hear this in reality and very clearly. And above all you do not hear the noiseshaping filtering.
ACOUSTIC INSTRUMENTS Most technicians and producers are experts who are used to dealing with pop music and digital musical instruments. The only -*test*-('") however by which a converter can be really evaluated on its merits is the recording and reproduction of acoustic instruments. When did you last visit a symphony concert? If the soundcharacter of the live performance of the symphony orchestra still resonates in your ears, then you know that SACD tries to fool your ears.
R.A.B. September 2000
Addendum: Direct eXtreme Digital SACD is a high capacity storage medium. DSD (Direct Stream Digital) is the format in which the data are stored on the SACD.
DSD is a 1-bit format, a so called pulse density format and definitely not a PCM format (Pulse Code Modulation).
DSD is suitable for storing high resolution analog tape recordings. Initially SACD was developped for storage of old analog tape recordings.
The signal is transferred without intervention into the DSD format: Direct Stream Digital.
DSD has two negative aspect:
1. Noise shaping is applied.
This deteriorates the pulse. The filter of the noise shaping is hanging like the sword of Damocles over the music signal.
2. Editing is impossible.
A digital format should have at least 2 bits to be able to edit the signal. DSD is a one bit encoding/decoding format, which means that it is not possible to edit in DSD. Only in a pulse width low bit format editing is possible.
In order to use the full benefit of the SACD with its Direct Stream Digital high resolution, and to have the possibility of editing the recording, it is of course necessary to make recordings with a high resolution medium like the analog tape recorder or any high resolution PCM (multibit) format.
After editing, the recording can be transferred to DSD and can be stored on the SACD.
But what high resolution PCM recording format? The format of the Compact Disc with its 16 bit and 44.1 kHz. or the 48 kHz. of the digital cassette recorder can hardly be called high resolution formats.
If editing is not possible in a 1-bit (pulse density) format, it is necessary to develop a format in which it is possible.
Now a new recording format has been developped: DXD, Direct eXtreme Digital.
DXD is a format in which editing is possible. It has a higher sampling
frequency and a higher resolution than the early digital format of 44.1 kHz. sampling frequency and 16 bit.
DXD is the low bit, pulse width format, which uses 5 bits.
The advantage is not only the possibility of editing. It also needs only about half the level of noise shaping which the DSD recording system needs.
The result is that the use of DXD means that the pulse has also improved. (You can imagine what the quality would be when working in a 16, 24, 32 or 64 bit format which do not need noiseshaping at all and give maximum pulse.)
Nevertheless the introdution of DXD seems to be good news. Now any recording made in whatever format can be converted to DXD and after editing the signal in DXD, it can be converted to DSD and stored on the SACD. The use of DXD means better sound coming from SACD.
Quote fromhttp://www.soundfountain.com/index.html