CD-4 stands for Compatible Discrete 4 Channel. This was a true four channel surround sound system for vinyl records. In America, the system was known as the RCA Quadradisc system. The system was launched in the early 1970s in Japan and a few years later in the USA.
Stereo Lab will support software decoding of CD-4 quadraphonic LPs. Feedback from our user-base has illustrated that interest remains in this format and a requirement for a software decoder exists. CD-4 (Quadradisc) has the 4th largest discography of commercial quad systems after: Dolby Surround; QS and SQ. The timescale for the integration in the software is not yet decided. However, the concept has been proved and we are ready to code this when the next development slot becomes available.
There exists an interesting parallel between the contemporaneous, rival, analogue television standards and the rival quadraphonic standards. Rather as with analogue television standards, a dimension exits which might be labelled quality vs. robustness when comparing the performance of the rival "quad" systems. Thus, a colour television standard like NTSC colour was capable of fine reproduction, yet it remained sensitive to channel distortions which made it a rather "fragile" system: whereas, the French SECAM system, whilst not capable of very fine reproduction, was so robust that colour TV signals could be passed (and even recorded) on monochrome equipment. Judged in this way, one can see that it's simply meaningless to declare one system as "best", since the question, "best for what?" follows inevitably.
Matrix quadraphonic systems like QS and SQ can be seen as the "SECAM of quad": clever, robust with the extra information hidden in the same bandwidth of the audio and undemanding of replay equipment: whereas JVC's CD-4, whilst certainly the only system capable of real, discrete 4-channel performance, was a relatively fragile quadraphonic system. A number of requirements both of the equipment and the medium need to be fulfilled to derive a stable and realistic quadraphonic effect; conditions which are even harder to satisfy half-a-century on from the system's heyday!
Because of this, and because we can't control the quality of the hardware needle-drops applied to the software decode, Stereo Lab is designed to "fail gracefully" when those conditions are not satisfied. To achieve this, the decoder in Stereo Sauce converts the CD-4 signal to an intermediate Ambisonics B-format signal before the final loudspeaker signals are calculated. To do this has a number of important advantages as explained below.
Baseband signals (20Hz - 15kHz) out of the groove are in fact the sum of the front and back signals (LF + LB) and (RF + RB). These signals are engraved on the 45° walls of the disc just as in a conventional stereo disc. The difference signals, used to separate back from front, are FM encoded on a pair of ultrasonic (30kHz) subcarriers above this baseband signal. (See diagram above.) The overall bandwidth of the CD-4 signal is thus from about 20Hz to 45kHz; a rather profligate three times greater than for conventional stereo LPs; a requirement which required CD-4 discs to be mastered at a third of normal speed.
A special (Shibata or hyper-elliptical) stylus is required to follow these high frequency subcarrier signals when the record is rotated at normal replay speed, and various precautions, such as: wide bandwidth preamps; correct cable types; low HF crosstalk in the cartridge and the subsequent electronics, are all required to recover subcarrier signals of sufficient and constant amplitude that successful decoding is possible. A micrograph of a recorded CD-4 groove illustrates the presence of the carriers as well as emphasising the concomitant requirement for scrupulously clean media and precision replay equipment.
Essentially, the CD-4 decoder in Stereo Lab has an Ambisonics "back-end".
This is included because Ambisonics is a formalised description of a sound-field, so it is possible to control the adaption of the decoding on a more elegant basis than simply collapsing the front-back separation in the event that the recovery of the FM signals is not entirely successful (which is all the CD-4 standard allows for).
Our approach has been to recover an enjoyable and satisfying musical effect from CD-4 records for final presentation using the standard ITU-R BS. 775 5.1 surround-sound loudspeaker layout.
It is hoped that, in this manner, the decoder will be generally useful to a wide range of users with different standards of replay equipment and that these adaptive techniques will help redress some of the hostility towards the CD-4 system where it has sometimes gained the epithets, sandpaper quad or more generally seedy four.
Importantly, ONLY needle-drops at 96kHz sampling are acceptable for CD-4 decode in Stereo Lab. All other sample rates simply produce an error ("wrong sample-rate") and fail to process.
Needle-drops must be be recorded without RIAA equalisation1. Attenuation of the upper sidebands of the CD-4 FM signals due to RIAA filtering is over 46dB relative to the baseband. Thus ninety-nine percent of the FM carrier level is lost in the RIAA filter and the FM tracks are reduced to 8-bit resolution. Our experiments have shown that reliable CD-4 FM carrier recovery is not possible with RIAA equalised needle-drops. (In any case, a non-equalised needle-drop is the usual input format for Stereo Lab phonograph processing and highly accurate RIAA correction to the baseband signals will be applied in software.)
Cartridge level matching is forced in CD-4 processing (where the value is derived from the carriers), so there is no need to select cartridge level matching in the PHONO preferences dialogue. No click-pop reduction systems are possible when decoding CD-4. Rumble filtering should always be selected as sub-sonic frequency components can upset the FM discriminator.
Note that successful CD-4 decodes require some special requirements of the needle-drop hardware:
It is highly recommended that you use audio software such as Audacity* to record CD-4 needle-drops to the Mac. This software offers an instant view of the audio waveform as the needle-drop is made and the presence (or not!) of the 30kHz carrier is easily witnessed and judged; most obviously in silent sections where the presence of the subcarriers and their channel balance is easily inspected.
Using Audacity to record CD-4 discs also enables the correct recording levels to be set. CD-4 discs are recorded substantially quieter than conventional stereo discs, so be careful not to under-record CD-4 needle-drops 6.
When looking at the waveform of an unequalised needle-drop (which is the only mode supported in Stereo Lab), the carrier should appear as a band of constant modulation.
The screen-grab below illustrates well the situation where, due to an incorrectly set skating-force adjustment, the carrier on the right channel (lower trace) is intermittent and inadequate for demodulation.
Once the turntable and tonearm setup is optimised, don't assume this will be remain the best set-up across the whole side of an CD-4 LP - the wavelength of the modulation in the inner grooves of the record is ⅓ of the wavelength of the modulation at the outer edge of the record, so just about every aspect of a CD-4 needle-drop becomes more critical as the record plays.
The state of cleanliness on CD-4 records is especially important. Cleaning CD-4 records is something of a specialist subject. We have found that standard methods should be considered a baseline for CD-4 types and that more Draconian methods are needed on discs which have been played many times.
If the quality of the ultrasonic subcarriers for either channels is judged inadequate, the CD-4 decoding process in Stereo Lab is halted and the warning "no RADAR" appears in the process progress dialogue. (RADAR is the term JVC used for their carrier detection in the heyday of CD-4, so we use the term too; although what RADAR stands for in this context is not known.)
Further processing is not possible if this warning is given.
* Audacity is a free, easy-to-use, multi-track audio editor and recorder which runs on the Mac platform (and, in fact, on Windows and GNU/Linux too). Audacity is free software, developed by a group of volunteers and distributed under the GNU General Public License (GPL).
The first step is to separate the "quad" signal so as to
1) derive the baseband signals and
2) separate the HF carriers for the quad effect.
This separation is achieved in linear-phase, 8-pole filters; an improvement over the original analogue filtering (shown right). The low-passed signals are treated as standard stereo signals.
The demodulators are the heart of the CD-4 decode and the demands on this element are severe in the CD-4 system as the message frequency range is half that of the carrier and the modulation index is over 30 at low frequencies. The graph left illustrates the modulation index (at 0VU) of the FM signal relative to frequency and how this is modified by the signal pre-emphasis prior to being applied to the FM modulator. (This is from a Service Manual of a CD-4 demodulator: the red-line, indicating a 1kHz deviation, is added for clarity2). Peak deviation at 1kHz modulation is ±7kHz on a 30kHz carrier5. Compare this with broadcast FM where a 10MHz IF is subject to ±75kHz deviation.
Provision must be made that the FM discriminator itself offers a high degree of AM rejection because carrier amplitude is modulated by a host of secondary effects in the CD-4 medium. (It's an illuminating experiment to AM demodulate the CD-4 carriers and listen to the result.) Limiting is universally used in analogue decoders to eliminate carrier amplitude changes, but such a simple technique is not appropriate in a time-discrete system unless the signal is massively oversampled. In the Stereo Lab software, a fast-acting AGC system is employed to maintain constant carrier level.
The FM discriminator design in Stereo Lab required considerable development. Not only does the FM signal require a very linear discriminator , but the algorithm must be relatively tolerant of carrier signal offset and drift due to the the speed variation and accuracy of the playback deck. It must also re-lock quickly in the event of carrier drop-out so that it does not produce noise like an FM radio when detuned from a station.
Below is a short video which illustrates the spectrum of a demodulated sweep at 0VU from 100Hz to 10kHz in the FM channels. The analysed signal is direct from the Stereo Lab demodulators so that you can also see the (unfiltered) sideband structure around the 30kHz carrier. Distortion due to demodulator non-linearity is a few percent (second and third harmonic).
With all these precautions, we have found that the quality of real-world demodulated signals from 50 year old records is considerably worse than the performance from the demodulation of our own test signals which leads us to the inevitable conclusion that considerable distortion is introduced in the encoding and replay of CD-4 discs. Why is this?
Theory suggest that the FM subcarrier system for CD-4 is capable of first-rate performance5. However the practical implementation of this system is bedevilled with issues which undermine the theoretical performance.
Any interference from the baseband channel (up-talk), not simply as a result of inadequate filtering, but due, for example, to harmonic distortion products from the baseband appearing in the FM band, are a source of particularly nasty, "splashy" distortion when demodulated. So is crosstalk between the two FM channels in any part of the system - including in the fragile cartridge. Most modern cartridges are unsuitable for first-rate performance because they offer inadequate channel separation at 30kHz. A few manage about 12dB separation, JVC stated that a minimum of -20dB was necessary for adequate performance of the CD-4 system. In our view, even -20dB is wide of the mark. Listen to the effect of 20dB interchannel crosstalk on a track with a sinewave sweep on one-channel only (the right side is silent). Apart from the obvious effect of crosstalk, the live-channel audio is badly distorted in the encounter: it sounds like something from the BBC Radiophonic Workshop!
The frequency spectrum graphic, captured as the pure tone passes the 100Hz mark illustrates the distortion vividly with a procession of distortion components above the 100Hz fundamental.
Wow and flutter are further sources of "extra" (and troublesome) frequency modulation.
All these unwelcome frequency components in the FM band result in a host of spurious cross-modulation products which create distortion and raise the noise-floor. It is for these reasons that the practical CD-4 system included an aggressive noise-reduction system (ANRS). 3, 4 In fact, it's no exagerration to say that CD-4 would not work at all were it not for this multi-band expander arrangement. Interestingly, the Marantz CD-400 demodulator manual does say this regarding ANRS,
ANRS is used in the CD-4 system to reduce interference and noise caused by crosstalk between the left and right modulated carriers..
In other words, the "noise reducer" isn't there to reduce channel noise so much as to suppress co-channel interference.
We incorporate both modern correlation-based noise reduction techniques and frequency-based noise-gating within the software decoder. We base these techniques on software emulation of four hardware decoder references; the first is based on the Marantz CD-400 (JVC 4DD-5)2 and two are based on the QSI-5022 integrated circuit decoder designed by Louis Dorren3, 4. Finally, the late JVC CD4-50 has had a great influence on the software design. The CD4-50 is a rare hardware decoder but is widely reckoned to be a decoder which behaves where others fail. This is due to the rather different operation of the ANRS circuits in this late implementation. We have incorporated these changes in the Stereo Lab software decoder.
The matrix section in a standard CD-4 hardware decoder simply sums the FM signals and the baseband signals to generate the forward pair of signals and subtracts them to generate the rear pair of signals.
Our alternative approach in Stereo Lab is to derive the three (B-format) signals of first order, horizontal Ambisonics from the four CD-4 component signals. This enables us to exploit the power of Ambisonics to decode to different loudspeaker layouts. Using this technique, the CD-4 decoder in Stereo Lab decodes CD-4 to a standard, modern ITU-R BS. 775 5.1 surround-sound loudspeaker layout.
Hardware decoders require adjustment of the FM/baseband sensitivity; usually via a couple of potentiometers on the rear of the CD-4 decoder case. These are adjusted using signals from a special test record which was supplied with the unit. In Stereo Lab, no such special material is required and this adjustment is made automatically.
A short demo of our present progress is available in the Phaedrus Audio listening room.
The input file to the CD-4 decoding process should be simple, two-channel, non-equalised, stereophonic track derived from the LP replay (or other source) at 96ksps ONLY. The output may be selected to be separate mono files (useful if you want to re-encode in DTS for example), or as a group of stereo files encoded:
Pspatial Audio have received much help and advice in the development of this decoder. Thank you to all who have contributed ideas and samples of CD-4 material. We should especially like to thank Kirk Bayne and Jonathan Gatarz for their invaluable help.
1. Tucked away in the original reference relating to CD-4 is the information that all the "extra" subcarrier information is recorded constant-velocity. The diagram below indicates this detail that the modulator equaliser compensates with a downward slope for the upward slope of the RIAA equaliser in cutting. So, the unequalised signal directly from a MM or MC cartridge has the correct frequency balance to ensure the upper and lower sidebands are correctly balanced. This is why an unequalised needle-drop is preferred.
A DISCREET FOUR-CHANNEL DISC AND ITS REPRODUCING SYSTEM (CD-4 SYSTEM), Inoue, T. et al. Presented at the 39th Convention of the AES, October 1970. The manuscript for the presentation was expanded and revised in March 1971 and appeared in the JAES July/August 1971 (title unchanged).
2. Maranz Model CD-400 Service Manual.
3. Build a High-Performance CD-4 Demodulator. Dorren, L. Popular Electronics September 1975 (illustration above)
4. Surround Sound Decoders 1 & 2. Heller, D. Wireless World June, July 1976
5. Recording characteristic of the CD-4 FM channels (from Reference 1). If performance from these channels really delivered 74dB dynamic range in a practical decoder, noise reduction would not be required.
6. Comparing the recording levels of standard stereo and CD-4 discs is complicated by inconsistent use of RMS and peak velocity values in the various references. Standard level for a normal, stereo LP is 3.54cm/s. This is an RMS figure and relates to a single stereo channel. If both channels are recorded at standard level, the resulting velocty is the vector sum of the two individual channel velocities. Because the channels are inscibed on two, perpendicular groove walls is equivalent to the root of the sum of the squares of the individual velocities (Pythagoras' theorem). Thus,
√(3.542 + 3.542) = 5cm/s.
It's easy to get this mixed up with the peak velocity which relates to the RMS value of 3.54 cm/s which is, of course,
√2 × 3.54cm/s = 5cm/s.
But these two values are quite different from each other.
The CD-4 "standard" (the nearest thing we have appears to be Ref. 1) sets standard recording level at 2.23cm/s probably RMS and for one channel. So, CD-4 level is,
2.23/3.54 = 0.63 , or 4dB below standard level on a stereo disc.
Each, unmodulated carrier is recorded at a velocity of 3.54cm/s and thus appears (because it appears before the baseband moduation) as a signal at exactly reference level when recording a non-equalised needle-drop. In fact, this is an excellent way to set recording-level for a CD-4 disc (provided it is known that the cartridge has an extended frequency response).
For all support issues, go here.
For Pspatial Audio sales, email: email@example.com
Apple Certified Developer. Stereo Lab, Aria 51, Aria 20, Head Space, Groove Sleuth, iLOOP and FRANCINSTIEN T-Sym are trademarks of Pspatial Audio. FRANCINSTIEN and Bride of FRANCINSTIEN (BoF) are trademarks of Phaedrus Audio. Macintosh and the Mac logo are trademarks of Apple Computer, Inc.