Minidisc FAQ: Audio Topics

  1. Is there an audible difference between modern MD and CD/DAT?

    Since MD stores audio using a data reduction technique (see ATRAC, below) there are considerable differences between the audio signal from a CD and that of an MD digitally recorded from it. Whether or not these differences are audible however depends to a large degree upon who is doing the listening, most MD users consider the sound quality of modern MD equipment to be essentially the same as CD. But even those who can hear differences usually find them to be undisturbing and inconsequential, frequently being unable to say which is which. (Note that a proper A/B comparison of the two formats necessitates using the same DAC. If an outboard DAC is unavailable, or the MD unit (such as a portable) does not provide digital out, the MD unit's monitor mode can be used to do D/A conversion of external digital signals, thereby allowing comparison with a common DAC.)

    A small double-blind test made by the ABX Company found that subjects did not find a difference between ATRAC processed music and its original, but could readily spot the difference when a special test signal was used for the comparison.

    However, one significant difference with DAT is that differences with the original increase in each subsequent copy generation, even when recording digitally. Each time the MD is played, a full 16-bit data stream is regenerated from the compressed data. Each time it records, the input data stream is recompressed. The artifacts of the compression process build up from generation to generation. See ATRAC below.

  2. Between MD and DCC?

    Yes. DCC uses a lower-loss compression algorithm than MD (4: 1 vs. 5: 1), and thus could sound better. MD should be compared to DCC with the same DAC, for fair results. Also note that most tests compared the first generation Philips home DCC recorder with the first generation Sony portable MD recorder (MZ-1).

  3. What is ATRAC exactly? How does it compare to PASC?

    ATRAC (Adaptive TRansform Acoustic Coding) divides the 16 bit 44.1 KHz digital signal into 52 sub-bands in the frequency domain (after a Fast Fourier Transform). The sub-bands in the low frequencies are finer than the ones in the high frequency range. A psycho-acoustic transfer function that takes advantage of the masking effect and the absolute hearing threshold then removes enough information to reduce the data stream to 1/5th of the original size. Each channel receives that treatment separately (the Sony MZ-1 portable MD recorder features one ATRAC encoder/decoder chip per channel). PASC (Precision Adaptive Sub-band Coding -- used in Philips now defunct DCC [Digital Compact Cassette]) divides the digital signal into equally spaced sub-bands and removes less information (to only 1/4th of the original size). PASC is essentially the MPEG Layer 1 audio standard (can be decompressed with MPEG Layer 1 players after a trivial preprocessing step).

    Both are data compression algorithms, used to store the information content from a stream of 16-bit samples in fewer bits. The purpose of compression is to reduce the rate at which the disk has to deliver or record bits, and to reduce the total number of bits stored. There are many compression algorithms. The ones used for computer data (for example in archiving programs) are lossless; the result of decompression is identical to the input.

    PASC and ATRAC are both "lossy" algorithms. In order to get greater compression, they do not attempt to preserve every bit of the original data, but rather only the acoustically "important" bits. Considerable cleverness goes into finding the sounds masked by properties of the human auditory system, ones that you would not hear even if they were reproduced. By all accounts the two schemes do amazing well, considering they operate in real time.

    See the AES paper on ATRAC for further technical details.

  4. What is the bit rate of the Minidisc's ATRAC audio after compression?

    For a stereo signal it's 292162.5 bits/sec. ATRAC compresses 512 incoming 16 bit samples (1024 bytes) into one ATRAC ``sound group'' (212 bytes) giving an audio compression ratio of 4.83:1. Here is the math:
    44100 samples/sec (incoming single channel rate)
    / 512 samples/soundgroup (giving 86.133 soundgroups/sec/channel)
    * 2 channels (giving 172.266 stereo soundgroups/sec)
    * 212 bytes/soundgroup (giving 36.5K stereo bytes/sec)
    * 8 bits/byte (giving stereo bits/sec)
    = 292162.5 bits/sec.
    ATRAC3 (used in MDLP) runs at 132kbps (LP2) and 66kbps (LP4). See the MDLP FAQ for further details.

  5. How does ATRAC compare with MPEG compression? At what bitrate would an MPEG file be equivalent to a song compressed with ATRAC?

    ATRAC is 292kbit/sec, giving ``CD like'' audio fidelity. MPEG Layer 1 (i.e. PASC) gives transparent CD fidelity audio at 384kbit/sec, Layer 2 (i.e. Musicam) and Layer 3 give ``CD like'' fidelity at 224kbit/sec and 128kbit/s respectively. A user has compared ATRAC and MPEG Layer 3 and rates ATRAC far better.

    VQF has a system offering 18:1 compression, pitched as something of an alternative to MP3.

  6. If I use digital connections between an MD player and MD recorder, why can't I make an exact copy of an MD?

    The compression of linear (PCM) digital audio into ATRAC format and expansion back again is an asymmetric process. Even if you are compressing and expanding data that has been through the compression and expansion cycle once before, the result of the operation will yield data that is slightly different from what you started with. This may seem strange, since fundamentally it could be done symmetrically. However, there are a few uncontrollable asymmetries in the process. One is the framing of input data. ATRAC starts by breaking the input signal into blocks of 512 samples. If it were possible to arrange that the sample block boundaries created during decompression be used when the signal was further recompressed, it would improve the symmetry of the process. Another is that ATRAC introduces noise, usually at an inaudible level, but in subsequent compression this noise is indistinguishable from signal and can alter what ATRAC determines to be masked, thereby changing the number of bits allocated to each band. Also, mathematical round off error in the compression and decompression calculations may introduce noise as well. In some MD copying experiments done out to 100 generations you can hear the artifacts of the compression process. The technical reasons for generational loss are examined in Frank Kurth's paper An Audio Codec for Multiple Generations Compression without Loss of Perceptual Quality (pdf).

    The next question then becomes, why not copy the compressed data directly, thereby avoiding the asymmetry of compression and decompression? Unfortunately, consumer grade MD machines do not provide access to their compressed data, nor do they provide a way of directly recording compressed data, even if it were available. The S/PDIF digital interconnect only carries data in the linear (PCM) format. Professional machines however (such as the Sony MDS-B5 with direct ATRAC I/O) do allow exact bit for bit MD copies to be made.

  7. What else (besides compression) affects the sound quality of MD?

    The entire rest of the sound reproduction chain is important. The digital to analog converter (DAC) is key to reproducing the sound from the decompressed data. The audio amplifier circuitry is also critical. And there have been reports of MD playback machines (not Sony) that were made unlistenable by poor-quality attached headphones!

    Every component of the chain from the DAC to the eardrum is important to good sound reproduction. All elements other than the compression algorithm must be held constant before A: B comparisons are made, for example.

  8. Does ATRAC have "forward compatibility," or is it a static algorithm? Does the ATRAC version affect the quality of the recording or the playback?

    The encoder (recording) side of ATRAC offers room for improvement from one generation to the next (specifically, in the decision about how to allocate encoding bits so as to best match human psycho-acoustic properties). Thus, MD recordings made on a newer machine with a better ATRAC encoder will sound better than old recordings, even when played back on an old machine.

    The decoder (playback) side of ATRAC has a fixed structure, and though ATRAC chips are all generally expected to decode with nearly the same quality, increases in digital signal processing accuracy may allow slight audio quality improvements (if those improvements have not already been made to modern ATRAC chips).

    Since the ATRAC encoder plays the largest role in how an MD sounds, the implication for making digital copies between two MD units is to use the older unit for playback and the newer one for recording. When making analog copies, the relative quality of the A/D and D/A converters must be born in mind.

  9. What's the difference between the various ATRAC generations, and how well do they interoperate?

    The following sections refer to Sony's ATRAC. Sharp and Matsushita (among others?) have developed ATRAC chips of their own, though rather little has been made public about their evolution. While all manufacturer's ATRAC versions interoperate and are based upon Sony's original specification, it is not possible to compare their version numbers. Version numbers frequently serve to enumerate Minidisc/DSP chip generations as much as actual ATRAC algorithm changes, so there is no direct relation between version number and audio quality.

    ATRAC General Characteristics

    Showing information about each particular ATRAC generation.

    (From the February '96 issue of the German Stereo magazine. The phrase sparkling noise in the text below is an imprecise translation of Spratzeln, which has been said to be typified by the sound of CO2 bubbles bursting on the surface of a beer).

    ATRAC 1 ATRAC 2 ATRAC 3 ATRAC 3.5
    Introduction 1993
    (Too early due to DCC)
    1994 1995 Current
    Threshold 15KHz 18KHz 18KHz 18KHz
    Noise Big! Much lower than ATRAC 1 Dynamic filter: no noise in breaks ==DAT
    Sound metallic close to DAT no difference to DAT in "blind listening test" ~=DAT
    Sparkling Noise hearable not much better only hearable in very silent passages gone

    ATRAC Interoperability Table

    Showing results of recording and playback between various ATRAC versions.

    ATRAC 1 recorder
    MDS-101
    ATRAC 2 recorder
    MDS-501
    ATRAC 3 recorder
    MDS-303
    ATRAC 3.5 recorder
    MDS-JA3ES
    ATRAC 1 Player (see table above) 15KHz! Less noise 15KHz! Less noise 15KHz! Less noise, "sparkling noise" gone
    ATRAC 2 Player 15KHz threshold! Less noise, still metallic sound (see table above) No difference from ATRAC 2 recorder No "sparkling noise", a bit less noise (as when recorded with 2.0)
    ATRAC 3 Player No big difference from above No difference from ATRAC 2 player (see table above) No "sparkling noise", a bit less noise (as when recorded with 3.0)
    ATRAC 3.5 Player No big difference from above "sparkling noise" still not better, but less noise than above low level "sparkling noise" remains, a bit less noise than above (see table above)

    In general, all ATRAC versions are fully compatible with each other. However, if you play or record something with 1.0 the result will be rather poor no matter from which version the source came from or goes to. If you take a higher version the result is generally be good enough, when you use ATRAC 3.5 for either playing or recording it gets even a bit better. For portables and car-players with 3.0, the weakest link is the A/D converter.

    The practical result: buy a 3.5 or better for 'home use', record there, and you get better quality in your 3.0 portable or car-player.

    Multigeneration Copy Tests

    Showing results of defeating SCMS copy protection and making multiple generation copies with each ATRAC generation.

    ATRAC 1 to ATRAC 1 After 5 generations unacceptable, after 20 generations awful.
    ATRAC 2 to ATRAC 2 After 5 generations no hearable difference, after 20 generations tiny distortion.
    ATRAC 3 to ATRAC 3 Not much better than with ATRAC 2.
    ATRAC 3.5 to ATRAC 3.5 Slightly better than ATRAC 3, relating to noise.

    Translated by Felix Gers.

  10. What actual changes have been made to ATRAC over the years?

    There is not much information about this other than what Sony says in their brochures and what appears in occasional magazine articles (thanks primarily to Japanese MJ and German Stereo magazines for their detailed reporting). Since ATRAC's transform window size (11.6ms) and signal processing structure are fixed, the improvements come primarily through improvements to the signal processing step's mathematical accuracy and refinements to the bit allocation process. In brief:

    ATRAC 116x16 bit multiplication, no short mode blocks generated by encoder
    ATRAC 2 and 316x24 bit multiplication
    ATRAC 3.5Block Floating type calculation to improve performance on small signals (see: Japanese MJ Magazine MDS-JA3ES review)
    ATRAC 424x24 bit multiplication, frequency response still 19kHz (see: German Stereo article on ATRAC 4)
    ATRAC 4.5Adaptive High Band Control, frequency response pushed to 20kHz, noise lowered 3dB through higher computation accuracy (see: German Stereo MDS-JA50ES review)
    ATRAC using Type-R DSPFaster DSP allows two pass bit allocation algorithm that looks for redundancy and makes better use of available bits, providing further improvements to high frequency performance. (see: Sony blurb on ATRAC using Type-R DSP). Not known to have any improvements to decoding.
    ATRAC using Type-S DSP ATRAC chip that combines an upgraded ATRAC3 codec (with apparently improved decoding performance) and an ATRAC1 Type-R codec. (see: Sony MDS-JB980 equipment table)

  11. Is there any loss of information when I record from a CD?

    There are two sources of distortion. One is the chain of components that brings the sound to the MD's input. If you go analog-to-analog, you introduce the CD's DAC and the MD's ADC chips, each with its own artifacts. However, you can bring the digital data stream directly to the MD; then the only source of differences is the ATRAC compression algorithm.

    The ATRAC encoder removes information from the audio material in order to store it on the MD (5: 1 compression with loss). To make better MD recordings from CD, connect the MD recorder to the CD player via a digital connection (if possible). Thus, the ADC (poor in the first generation machines) cannot affect the sound quality. Otherwise, when recording via the analog input, make sure to adjust the manual recording level on the MD machine so that the meter peaks just above -12 dB (on the Sony MZ-1, never enable the AGC for CD recording).

  12. When I record from tape, microphone, or other analog sources?

    Yes, because the MD's analog to digital conversion circuits are involved, in addition to the ATRAC compression. High-end MD decks frequently have sophisticated analog circuitry, offering improved recording and playback performance over mid-range decks even when they share the same ATRAC chip.

  13. I don't have the equipment to make a digital recording, will my recordings sound okay?

    Providing you've got a clean source signal, analog recordings generally sound fine, giving nearly inaudible differences from digital ones. Note however that many computer soundcards have noisy analog output stages, so when recording from a computer, digital transfers are recommended (one exception is the USB based Xitel MD-Port AN1 which runs outside of the PC chassis and produces good quiet analog output). When making analog recordings, do set the recording level manually to avoid the audible effects of the Automatic Gain Control (AGC) circuitry adjusting the level during very loud and very quiet passages.

    One advantage of analog recordings is that MD recorders flag them "SCMS-penultimate", meaning that a digital copy can be made of them. One disadvantage of analog recordings is that track marks will be laid down based upon the recorder's detection of silences in the analog source (depending upon recorder settings and capabilities); this is less reliable than using the digital indications of track changes available in an S/PDIF (digital) signal from CD players.

  14. Many have suggested setting levels manually when recording from analog sources. Any tips for how to best determine the correct level using the MZ-R3's and -R30's feeble little LCD bars?

    You will have the correct recording level when the level meter is just between 4 and 5 bars. Verified with an MDS-503 from Sony, between 4 and 5 bars on the MZ-R3 record level meter will give you somewhere between -3 dB and 0 dB.

    Regarding the 'R30: A user connected a 303 to the R30 using a POC (optical) cable and compared the two meters. The result: Just consider the top bar to be 'digital over' and try to adjust your level in such a way that the second bar doesn't light up too often when recording from an analogue source (even less when recording live). That should do it. Compared to the MZ-1 the R30 meter is not very useful.

  15. Is it worth it to get the equipment necessary to make digital recordings?

    Digital recording provides the most convenience when copying CDs: no recording levels need to be set, track marks are copied from the CD perfectly, and analog to digital conversion artifacts (real or imagined) are completely avoided. The one problem with digital recording is that SCMS will prevent further digital copies to be made from the copied MD.

    Crutchfield's Tip of the Week is devoted to analog vs. digital recording to Minidisc.

  16. Which ATRAC chips are in which MD units?

    This information came in part from a Japanese magazine "MJ" and the 9/96 issue of the German Stereo magazine:

    IC Generation IC Part Number MD Deck Introduction Date
    ATRAC 1 CXD-2527 MDS-101 2/93
    ATRAC 2 CXD-2531 MDS-102
    MDS-501
    11/93
    2/94
    ATRAC 2 CXD-2531R MDM111
    MDH-10
    ATRAC ? CXD-2535CR-1 MZ-E40/E20
    ATRAC 3 CXD-2536R MZ-R3 5/95
    ATRAC 3.5 CXD-2536A MDS-JA3ES 6/95
    ATRAC 3.5 CXD-2536B MDS-503 10/95
    ATRAC 3.5 CXD-2536AR MZ-E40 1/97
    ATRAC 4.0 CXD-2650R MDS-JE500/S37
    N. American MDS-JE510
    MDS-JE700
    8/96
    ATRAC 4.0 CXD-2652R MZ-R30 8/96
    ATRAC 4.0 CXD-2652AR MZ-R50/MZ-R55/MZ-R37
    European MDS-JE510
    8/97
    ATRAC 4.5 CXD-2537R MDS-JA50ES 12/96
    ATRAC 4.5 CXD-2654R MDS-JE520 9/98
    ATRAC Type-R DSP CXD-2654R MDS-JA20ES/JA22ES 5/98
    ATRAC Type-R DSP CXD-2662R MDS-JB940, MXD-D5C 2000
    ATRAC ? (>= 4.0) CXD-2660GA MZ-R90/R91 10/99
    ATRAC ? (>= 4.0) CXD-2671-201GA MZ-R900 10/00
    ATRAC ? (>= 4.0) CXD-2671-202GA MZ-E900 10/00
    ATRAC Type-S DSP CXD-2664 MDS-JB980, MDS-JB780 8/02
    ATRAC ? CXD-2655R Grundig MD-P1

    See also the ATRAC version table.

  17. I've heard some negative comments about optical digital interconnects. Does is matter whether I use optical or coaxial digital input when recording to Minidisc? Is jitter a problem?

    In a word: No. The perceived problems with optical interconnects relate to an optical cable's greater theoretical potential to distort the digital signal, particularly to create small inaccuracies in the arrival time of data bits ("jitter"). However, in Minidisc recording jitter is not an issue since the digital input signal's sample values are recovered and passed directly into a memory buffer or into a sampling rate converter that is clocked with the clock embedded in the input signal. The sampling rate converter and/or memory buffer allow the audio samples to be subsequently read and passed to the ATRAC converter with an accuracy determined by the MD unit's internal quartz clock. Even if jitter was an issue however, it is doubtful that the short cable lengths involved in home HiFi systems could produce audible differences between optical and digital cables.

    Regarding the occurrence of outright bit errors due to a marginal cable: S/PDIF contains only parity information, there is no error correction capability. If the errors are bad enough to cause bits to arrive with incorrect values, the likely result is that the digital audio receiver will not be able to lock on to the signal.

    A short paper by DJ Greaves goes into further detail about S/PDIF, and has some comments about why jitter is not a problem even in equipment without buffers. Another paper by Tomi Engdahl goes into great detail about S/PDIF, even giving schematics for AES/EBU <-> S/PDIF conversion. Finally, Digital Domain has written a very comprehensive paper on jitter in digital audio systems.

  18. What's the scoop on digital connectors, and how can I make a digital connection between my source device with its digital coaxial output connector and my MD unit with its digital optical input connector?

    The short answer: If you are trying to make a digital connection between devices with differing signal types, you will need a converter. Core Sound sells something called a "Digital Format Translator" for $95 which will convert between coaxial SPDIF and TOS-link optical. There is more information on Core Sound's DFT page.

    A cheaper route, if you're willing to do a little electronics work, is to follow Shawn Lin's instructions for making a converter from parts.

    What follows is a discussion of audio digital interfaces. These interfaces come in 2 classes, optical and electrical.

    The optical format has two connector types: the small, squarish "TOS-link" connector and the optical miniplug, which has the same connector dimensions as a normal (electrical) mini-plug. You can buy optical cables with any combination of these two [male] connectors at the ends. TOS-link is usually limited to maximum cable lengths of 10 to 15 meters. The Sony part numbers for the optical cables are as follows: miniplug/miniplug: POC152HG, miniplug/TOS-link: POC151HG, TOS-link/TOS-link: POC-15HG. These cables can be ordered from Sony Parts (see below).

    There are two electrical formats. [the following excerpted from the DAT-link manual]

    SPDIF: (Sony/Philips Digital InterFace): This is the interconnect that is most often used on consumer DAT machines. The connectors are standard RCA phono connectors. This type of connector may also be lableled "IEC Type II" or simply "Digital I/O". Standard analog phono cables can usually be used for the digital data, however some cables that are designed for analog may not be able to carry the high rates needed for the digital data, especially over long distances. Many high-end audio stores carry special digital phono cables that solve this problem. [The pro-audio FAQ says not to use audio cables, but that video cables will work].

    AES/EBU: This type of cabling is most often found on professional equipment. It uses three-pin XLR connectors. Cables designed for analog applications work fine for AES/EBU connections as well. However, note that shielded cables (most cables are shielded) must be used, otherwise unacceptable levels of radio or TV interference may be generated. This type of cabling is the preferred choice for long distance runs between digital audio equipment.

    It is important to realize that there are subtle differences in the control information that is sent along with the audio data on these different connectors. The two main formats of this information can be broadly categorized into Consumer and Professional. For most applications, if you are using the SPDIF or fiber-optic connections, the consumer format applies. For AES/EBU connections, the professional format applies. Some DAT machines will not operate at all unless the correct format is used.

  19. I see MD equipment with 16, 20 and even 1 bit DACs (Digital to Analog Converters), what's the difference?

    A 1 bit DAC is more correctly referred to as a Pulse Width Modulation DAC.

    A conventional 16/20/... bit DAC uses resistive dividers to add a value proportional to the bit significance of each bit to its output voltage. As more bits are added to the DACs resolution, the more significant bits' accuracy must be improved to at least the value of the least significant bit, or there is no point in increasing the resolution. It is quite difficult to make a resistive divider network with the required accuracy.

    With a 1 bit DAC, the output voltage is produced by pulse width modulating a single fixed voltage. The accuracy is determined by the stability of the clock that times the width of the pulses - it is not difficult to very accurately time duty cycles using a clocked counter. All that is needed to increase the resolution of a 1 bit DAC is a faster clock and a counter with more binary digits. -Colin Burchall

    There are two classes of DACs. One-bit DACs and multi-bit DACs. A 20 bit DAC is (theoretically) better than a 16 or a 8 bit DAC. You can't compare them with one-bit DACs because they use another principle to convert from digital to analog. Multi-bit DACs always convert different values for the same time period. One-bit DACs convert the same value with different periods of time.

    Multi-bit DAC: Fixed time period, varying voltage/current.
    One-bit DAC: Fixed voltage/current, varying time period. -Ralf Kuchenhart

  20. Dolby ProLogic Surround is encoded in the stereo signal through phase shifting. Does the MD's transform coding interfere with the "Surround" information after recording and playback?

    I have done extensive testing with this, and have yet to find a single instance where the MD encoding/decoding affected the Pro-Logic encoding in any way. In every test I did, the surround steering information remained intact, and the result sounded precisely like the original.
    Scott MacLean ([email protected])


Copying Status: Public Domain. However, no liability is accepted for the correctness and safety any procedures or advice given here. This FAQ is distributed in the hope that it will be useful. Also, please include a notice that this file can be found at http://www.minidisc.org/minidisc_faq.html in any copies that you post.

Return to the MiniDisc Community Page.

As always, I would be much obliged for any updates and corrections from Sony Engineering. -Eric Woudenberg

Return to the Minidisc Community Page