CUTTING MANUFACTURING costs often results in inferior quality, which is exactly what happened when makers of compact disk (CD) players decided to use one digital-to analog converter (DAC) between the right and left channels. As a result, many CDs have an 11usec time delay between the left and right channel outputs. Some players, such as the Sony CDP-701, have two DACs, but most have only one and so suffer from the time-delay problem.
How the Time Delay Occurs
CD players use a 44.1kHz sampling rate, so the data for each channel has a new value every 22 usec. A shared DAC must process a new value every 22 usec. Unless a time correction is applied to the analog signal after the DAC reconstructs it, the right channel output will be 11 usec behind the left channel.
A CD player's outputs should be coincident when the digital data on the CD is coincident. Although no standard time displacement has ever been defined, devices such as the Sony CDP-101, CDP-200, CDP-400 and CDP-610ES and the Technics SL P10 have approximately 11 usec of interchannel time delay. An approximate correction (within 2usec) is used in the Technics SL-P7 and SL P8. Other players that use 2:1 over sampling and shared DACs have a 5.5usec interchannel delay. Most players that use 4:1 oversampling do not have a shared DAC or interchannel delay. Although delay can vary among makes and models, published tests often ignore this fact. Some times test results are erroneously ascribed to overall phase shift at high frequencies.
You can easily test your player for this problem. CD test records, such as Elektra's Digital Domain (9-60303 2), described in TAA 1/84 (p. 2), are available. To test for the problem, play track 19 (1kHz square wave) while monitoring both channel outputs with a dual-trace scope. Adjust the oscilloscope time base to display a single square wave and synchronize it on the left channel output of the CD player. If the player has an inter channel time delay, a displacement of the trace for the right channel will be visible. Figure 1 shows the results of this test on a CDP-101.
PHOTO 1: The author's phase corrector is housed in a die-cast box painted
black and labeled with dry transfer lettering.
Why worry about an 11 usec inter-channel delay? Sound travels only about % inch in 1llusec, and time delays of this order are common in analog tape systems. Test tapes used to align tape recorder azimuth are not always consistent within 11usec. If a tape recorder is aligned with one tape and checked with another, a discrepancy of 11 usec might occur. In fact, azimuth alignment errors in common cassette recorders often create time delays much greater than 11 usec. But the CD is supposed to be a lot better than cassette tape or any other form of analog recording, so an interchannel time delay of 1lusec that is reliably detectable in listening tests is significant.
Reliable Detection
One way to detect interchannel time delay is to listen to a time-delay circuit being switched in and out of one channel. (Switching a high-quality circuit in and out of both channels might not be reliably detectable.) If the time-delay circuit is reliably detectable, a time-correction circuit is needed.
A relatively simple all-pass filter can do the job. After reviewing several standard filter designs, I decided to go with one recommended by Stan Lipshitz (Fig. 2). It is composed of a buffer and Bessel-tuned (Q=0.5773) second-order all-pass filter. The buffer provides a stable source impedance for the filter, whose input impedance varies significantly with frequency. The filter has unity amplitude gain at extreme frequencies and can be tuned for unity gain at 48kHz to provide extremely flat response several octaves on either side of the 20Hz to 20kHz range. It provides a delay that is accurate within 1 per cent to 20kHz, which is probably more accurate than most players' channel matching. With readily available op amps, its noise and distortion performance matches or exceeds the CD players themselves. To test the corrector, I used the same setup as in Fig. 1, but I switched the corrector into the signal path. Figure 3 shows the results of my test.
Switch-selectable modes marked “record” and “playback” increase the corrector's flexibility. Playback is used with CD players. Record is used with EIA] standard digital audio adapters for video-cassette recorders, such as Sony's PCM-F1, when you must synchronize the digital data on the tape. This is desirable when ultimate playback is not by means of an EIAJ digital audio adapter, but by direct digital encoding onto a CD master.
This mode of transcription is al ready partially supported by switch able de-emphasis circuits in CD players. The interchannel time-delay issue is usually ignored. Normally, interchannel time delay during re cording of EIAj digital tapes is compensated for during playback through the digital adapter. If an EIAj digital tape is made through the corrector in the record mode, it should be played back through the corrector in the playback mode.
Since the essence of the corrector circuit is a simple time delay, you can adapt it for nondigital applications where small amounts of delay are required for both channels, such as in loudspeakers. For example, you can change the time delay to 5usec by changing C5-C8 to 470pF, or you can create a 1.1 msec delay by using 0.1 uF caps. Such a large delay in one stage will be accurate only to 200Hz, but it is suitable for electrically “moving ” a subwoofer about a foot farther from the listener. A subwoofer with an electronic crossover and its own amplifier could then be located near the listener to reduce standing-wave effects in the room. Several cascaded stages of delay installed between the crossover and the amplifier could synchronize the subwoofer with loudspeakers some distance away.
Similarly, 0.005 uF caps would produce a 50usec delay accurate to 4kHz. This electrically 'moves' a midrange driver 1 centimeter away from the listener. Several cascaded stages of such a delay would synchronize it with a woofer mounted on the same baffle board. Linkwitz came up with the idea of using all-pass filters to electrically synchronize loud speakers that are mounted with non coincident acoustic centers to meet other requirements.
FIGURE 1: Left and right channel outputs of a commercial CD player with a multiplexed D-to-A converter. Signal frequency is 1,002.273Hz. (Photo court. of DLC Design, Farmington, MI.)
With the 5532 op-amps, you should have no problem cascading this circuit more than a dozen times.
In cascaded applications, the second and later stages do not require buffer IC1, and only the last stage needs output coupling components C9, C10 and R13-R16. You can omit S1.
The parts list includes a note about using 1 percent resistors to eliminate potentiometers R3 and R4.
As recommended by D'Appolito (Speaker Builder 4/83, p. 34), you can change this circuit's Q to 0.707 for use in Linkwitz-Riley crossovers with three or more bands. The changed parts are as follows:
R5, R6-56k; R7, R8-22.3k; R9, R10--12.4k; R11, R12--31.6k; R17, R18--68k.
When C5-C8 are 0.01uF, the design frequency is 1,000Hz and sizing for other frequencies is as described above. It is important to use 1 percent resistors and matched capacitors when flat response above the design frequency is required.
The corrector bypasses the time delay by shorting out RC network R7-C5-C7 or R8-C6-C8. When both RC networks are shorted, the corrector is a unity-gain buffer. When neither RC network is shorted, an 11 usec time delay is applied to each channel. Since the program material has already been delayed weeks or months, adding an 11 usec delay has no impact on sonic accuracy, pro vided both channels are processed in the same way. You can use the two bypass modes to determine whether the corrector has any sonic defects, since neither introduces interchannel delay. The latter mode is available if you use a center-off switch.
Prototype Construction
I constructed my prototype unit with “analog wire-wrap” techniques. I built the power supply and filters on separate perforated boards. Although I used LF353 op amps for my initial check, I later installed NE5532 op amps for best noise and high-output performance. Figures 4 and 5 show the foil and component sides of the etched circuit board, which was designed by Alpha Electronics.
--------------
PARTS LIST
C10 4.7 uF, 35V electrolytic
0.001 uF, 50V, 10% polystyrene, polypropylene or polyester
0.1uF, 25V ceramic NES5532 or LF353 dual op amp 1/4-inch phono jack ct, C12 IC1, IC2 J1-J4 R1, R2, R13, R14 R3,R4 RS, R6 R7,R8 R9, R10 47k, 1/4-W, 5% 5k multiturn preset pot 10k, 1/4-W, 5% 6.8k, 1/4W, 5% 6.2k, 1/4W, 5% R11, R12 22k, 1/4-W, 5% R15, R16 2200, W, 5%
$1 DPDT on/off/on toggle switch Chassis, wire, mounting hardware, power supply (Old Colony).
Note: The preferred E96 values for critical parts are as follows: R5, R6-11.3k; R7, R8-6.49k; R9, R10-6.04k; R11, R12-2.21k; R17, R18-15k. If you use these 1 percent resistors and match C5-C8 to within 1 per cent, you may replace R3 and R4 with jumpers at a net savings in cost.
----------------
Initially, I measured the power supply voltage, frequency response, noise and distortion. I used a dual-trace scope to evaluate interchannel time correction. I adjusted the potentiometers for unity gain at the 180-degree phase shift, which should be 48kHz, +1kHz. Once adjusted, the unit had a frequency response within 0.1dB from 20Hz to 20kHz. If you cannot obtain this response, try increasing the resistance of R9 and R10 by adding 3900 resistors in series. Since the CDP-101 already has LF353s and NE5532s in the analog signal path, they are reasonable choices for further processing. With LF353s, distortion measured below 0.02 percent at 2V output, 20Hz to 20kHz, while noise was down at least 90dB. Distortion was difficult to measure with NE5532s, and noise was down at least 100dB. NE5532s also have more gain at the high end of the audio band and help op-amp circuits that depend on high gain-such as phase shifters and state-variable filters-perform closer to design specs.
After I completed layout and drilling, I painted the exterior of a Bud CU-234 die-cast box with flat black Krylon. I then applied dry transfer lettering and Varathane spray polyurethane varnish to produce a high gloss and protect the lettering (Photo 1). 1 baked the undercoat for 60 minutes at 200 degrees and the finish coat for 90 minutes.
When I installed the circuit boards (Photo 2), 1 repeated my initial checks. I then connected the unit to a CD player and evaluated its performance with an oscilloscope and the test CD. Finally, I installed the CD player and compensator in my audio system for listening tests.
Time-Delay Tests
I performed double-blind listening tests with an ABX CS-5 control sys tem and RM-2 relay module, al though an RM-1 would suffice. I matched the level in the system using a Heath IM-2260 digital volt meter connected to the output of the preamplifier, as shown in Fig. 6.
FIGURE 3: Left and right channel out put of the same CD player as in Fig. 1 with an all-pass delay network delaying the leading signal. (Photo courtesy of DLC Design, Farmington, MI.)
FIGURE 4: Circuit board for the phase-corrector circuit.
FIGURE 5: Stuffing guide for the phase-corrector circuit.
The board is fastened to the front panel by the stem of the switch and stand-offs at the corners.
The test was essentially a straight wire bypass test, but since the corrector does not have precise unity gain, the “straight wire” was a precision attenuator using paralleled 10 k-Ohm dual-wiper Alps potentiometers for each channel. Since the output volt age at 1kHz was matched to four digits, level match was within 0.01dB. I used a Philips test CD to ensure that frequency response in the listening situation matched the test bench performance previously ob served. This is not the case with every piece of equipment, and it can be risky to assume level matching in the audio band in a practical application, even after excellent test-bench measurements. The asymmetry be tween standard test-bench conditions and operation in a practical sys tem explains many situations where equipment measures the same, but sounds different.
Before performing the listening tests, I determined what form audible effects might take. The 11 usec delay was too short to cause significant variations in the positioning of apparent sound sources in a stereo sound stage. It is close, however, so that was one possibility. In the case of a pan-potted mono source, if one channel is delayed 11 usec and then combined with the un-delayed channel, a 0.5dB loss at 10kHz and a 3dB loss at 20kHz will result. This effect is shown in Table 1 (column 3) and Fig 7.
To determine whether the 11usec interchannel time delay was detect able, 1 performed the double-blind tests by switching the corrector in and out of the signal path and listening for differences. Testing the bypass modes showed no reliably detectable sonic defects in the corrector.
I used the corrector in the record mode to test delay detection in situations where it did not already exist.
In double-blind tests using mono phonic pink noise and listening to the channels electrically summed after one channel was delayed, I reliably detected the effect in every case. I also reliably detected the effect with 1kHz square waves. In stereo headphone listening, however, I could not reliably detect the effect.
Since the record mode provides a time delay that is opposite that required by CD players, you can use it to make the situation worse than normal. This helps in training listeners to hear the effect. Once trained, listeners can progress to comparing no compensation or correct compensation, which is more difficult to identify correctly.
-----------
PHOTO 2: The circuit board for Alpha's kit and assembled units has an attached regulated power supply, shown on the right.
TABLE 1: RIGHT CHANNEL PHASE LAG. Sum and difference errors due to an 11 usec delay.
FIGURE 6: Setup used for listening tests of the time-delay corrector.
The remainder of the audio system includes custom equalization, an electronic
crossover, two modified Dyna 400 amps, three Ohm Acoustics Model F loudspeakers,
and a Cerwin-Vega Model 189E 18-inch subwoofer in a 14-cubic foot Thiele-aligned
enclosure.
--------------
When the difference between the two channels is developed by an electrical subtraction circuit, as is done in ambience recovery and some four channel decoders, different kinds of effects result, depending on the program material. When the left and right channels have exactly opposite phase, the L-R signal has the same frequency response aberrations as the L +R channel when the channels are in phase (Fig. 7). When the left and right channels are in phase, a false difference is created, as shown in Table 1 (column 4) and Fig. 8. My listening tests involving electrical subtraction of the left and right channels with and without the corrector usually resulted in reliable detection.
By the way, the reliable performance of CD players seems to be encouraging a resurgence of interest in various matrixing schemes, which will work better when the left and right channels are synchronized.
Electrical subtraction is also inherent in FM multiplex transmission. In this case, the L-R signal is rematrixed before listening, so it is not heard directly. False differences increase dynamic range requirements on the subcarrier path, how ever, which is undesirable.
Source of Frustration
A common source of frustration in listening tests is that not all recordings will produce a difference that is reliably detectable in some recordings. I spent weeks listening for the effect to appear more clearly. Since the listening tests were double blind, it was possible to establish those cases where the delay was reliably detectable. When it was not, the out come of the test was random guessing. The setup precluded prejudice, as you cannot perform a carefully controlled test and obtain a positive result by any means other than actually hearing a difference. Also, when conveying results to my cohorts, I easily convinced them of the reliability of the tests because the statistical analysis indicated only about one chance in a hundred that the results were due to guessing.
Of all the listening tests, those done on my living-room system (which has a center-channel loud speaker that plays an electrically summed signal) seem most relevant.
In that environment, 11usec of inter channel time correction was notice able. I was able to determine the correct identity of randomly selected signal processing, either corrected or uncorrected, about two-thirds of the time. Placing this result in the perspective of other double-blind tests, time correcting a CD player is less frequently detectable than 2 per cent harmonic distortion, the sound of an amplifier that does not recover from clipping properly, or the effect of increasing the power output of an amplifier from 64W to 324W.
It is more frequently detectable than 1,200 degrees or less phase shift at 1kHz applied to both channels, up grading 20 feet of loudspeaker wire from #16 to #12, increasing the slew rate of a 200W per channel power amplifier from 12V per microsecond to 40V, or upgrading the coupling capacitors in a preamp from electrolytic to Mylar.
It turned out that the effect of interchannel time delay was most noticeable as a loss of extreme highs, those with which I associate the color white. It was possible to detect the correction by listening for the “most white-sounding ” extreme treble.
The effect was audible on several musical selections, including those where the acoustic image of a tambourine or wire-brushed cymbals was near or substantially off center.
The somewhat contrived test procedure, which included constant re playing of a critical passage from 00:14 to 00:17 on one cut, is justified because it produced results in a relatively short time.
I had to work hard to get reliable results in a few hours of listening, but it is likely that a reliable perception of improved performance would come after longer periods of relaxed listening. I'm also pushing 40, and my ears suffer from a tour of military service, so I'm sure those who are younger or whose ears have led a more sheltered existence could hear the effect much more easily and un der a greater variety of circumstances.
FIGURE 8: Response of differenced outputs of a CD player for a signal
that is applied equally to both channels, with an 11 u-sec interchannel
time delay. This assumes that the inputs to the differencing circuit
were balanced with a true mono signal.
Comparison Shopping
This device is not comparable to commercially available anti-aliasing filter phase correctors, whose audible effects are controversial. This controversy stems from the phase shift applied to both channels at high frequencies by analog sharp-cutoff Some designers have devised “correctors ” for this “problem, ” but have not provided reliable evidence of improved sound quality. Various kinds of analog filters are used for digital encoding and decoding, but the phase-shift “correctors” assume only one filter design and might be damaging in some cases. Some CD players use oversampling and digital filters to alleviate phase shift. Al though a few reviewers note a sonic improvement due to the reduced phase shift, many “scientific” listening tests have not found reliable results above 1kHz.
It is difficult to second-guess the cause of an aberration someone says he or she hears, but many effects known to be reliably detectable are present in casual listening tests. In addition to reducing phase shift, oversampling reduces or eliminates the interchannel time delay and affects frequency response. Many tests of oversampled versus standard CD players also involve the CDP-101, which has an inherent channel-level match problem that is only partially corrected in production units. Errors in the 0.3 to 0.5dB range are common. Channel-balance errors of this size are reliably detectable. Inter channel time delay is likely to be the second most frequently detected effect, followed by minor variations in frequency response. A significant benefit of oversampling and digital filtering is consistent frequency response.
Most of the CD players with analog filters I have tested have unit-to unit variations of around 0.5dB, which is reliably detectable in listening tests. Sample variations are often larger than the difference between the averages of two different models.
To put these minor variations in perspective, remember that it is common to have an analog vinyl record playback system with response that varies as much as 0.5dB per revolution. This is due to minor warps affecting the alignment of the cartridge's magnetic system. Many cartridges change their response even more with normal changes in room temperature.
After testing this circuit for several months, I have found it a worthwhile upgrade to my system. Many FM radio stations have started to use CD players with interchannel time delays, and I am sure that they and their listeners would be particularly happy with this circuit's benefits.
- Arnold Krueger
SUPPLIERS
Alpha Electronics PO Box 15026 Detroit, MI 48215 (circuit boards, kits and complete units including UL-approved power supply)
Digi-Key Electronics Highway 32 South PO Box 677 Thief River Falls, MN 56701 (all parts) Jameco Electronics 1355 Shoreway Blvd.
Belmont, CA 92701 (all parts) Mouser Electronics 11433 Woodside Ave.
Santee, CA 92071 (all parts except NE5532 and LF353 op amps)
Westcomp Electronics 37387 Ford Rd. Westland, MI 48185 (all parts)
-------------------
Also see:
A DIGITAL NOISE GENERATOR, By Benjamin L. Poehland