Hello to everyone.
This is a review and test of the Sony SCD-555ES SACD and CD player. This is an European and Japan model. It has been released at the end of 2000 and cost around 1,830 euros at the time (about $1,650 of 2000). It has no US equivalent.
Part I: Presentation
The Sony SCD-555ES is a stereo only SACD and CD player, without any digital output for DSD data, only for 16 bits @ 44.1 ksps PCM data. There is one set of RCA (unbalanced) output and a front phone output jack with volume control. The Sony offers 5 different digital interpolation filters for CD data to choose from.
This player belongs to the very first generation of SACD players from Sony. It is a direct hair of the SCD-1, the very first SACD player, and the almost identical SCD-777ES. In fact, the SCD-555ES is the last Sony player to have used Sony's own "Pulse Length Modulation" (PLM) digital to analogue conversion system that had been introduced in 1989 in the seminal DAS-R1a. NTTY has already reviewed two other CD players that use a Sony DAC chip of the same family: the CDP-X333ES and CDP-X559ES. Please go read the review of the former, where the Sony system is discussed in some length.
Those of you that are interested by technical design would be happy to know that the service manual of the SCD-555ES is readily available and, above all, that the theory of operation of the first generation of Sony SACD players has been very comprehensively described in a patent granted in 2001. As I have always found the ideas herein described very neat, I cannot resist to elaborate a bit on some core aspects of the design. The main goal of Sony was to get a conversion system that has as much commonalities as possible between CD and SACD replay circuits. Actually, it is not an exaggeration to say that Sony's first SACD players were actually regular Sony CD players that had somehow been hacked to also play DSD programs with minimal changes.
The heart of the player is the Sony CXD8762 digital signal processor (figure 1):
This chip takes the serial interleaved left/right PCM input (which can goes up to 24 bits depth, although only 16 bits are used in this player, no re-quantization being made inside the DSP), it de-interlaces the left and right data-stream in a parallel format and performs 8 times digital interpolation (1) to increase the sample rate to 352.8 kHz. Sony advertises some exotic optional "slow roll-off" filters that are user selectable. Apart from the usual high frequency roll-off associated with that kind of filters, the most interesting aspect of at least some of them is that the interpolation is not done in successive cascaded FIR filters, but in a single-stage FIR filter thanks to the processing power available on-chip. A standard interpolation filter based on a 3-stages FIR filter is nevertheless provided. After this first interpolation, a further 8 times oversampling is performed by a (2) linear interpolator (i.e. the computed additional samples follow a straight line from one original sample to the next original sample). A this point, the sample rate becomes 2.8224 MHz. Not coincidentally, this sample rate is the same as DSD. Then, the digital data are re-quantized to 15 levels coded with 4 bits in a 2-complement forms by the (3) noise-shaper (probably third order, as many other Sony chips are). The (4) PWM block is an actual digital to analogue converter which gives the option to use the chip as an integrated digital processor and DAC, but this converter stage is not used in the Sony SCD-555ES. The on-chip converter is by-passed by (5) solid-state logic switches and the 4 bits data are put out in a parallel format at the (6) stereo outputs of the chip. This switches also select the data coming from the DSD signal path. Incidentally, when this chip has been used as an integrated digital processor and DAC in some players, Sony gave it a different part number, CXD9556, but the chips bearing both designations are actually identical.
In the SCD-555ES, the D/A conversion is done in a following chip, the CXD9521. The conversion is a double differential method, i.e. the digital data are converted by four D/A working simultaneously to produce complement and inverted signals that are successively differentially subtracted in the analogue domain. The output waveform of one pair of D/A stage is as shown on fig. 2 (the other pair shapes the inverse of the shown waveforms). T corresponds to the period of the master converter clock, which is 45.1584 MHz, 16 times the rate of the noise-shaped data.
The (1) first D/A of the differential pair converts the 4 bits input data in pulse length proportional to the input data, the (2) second D/A of the pair produces a pulse length that is the complement of the the first one and the two pulses are (3) differentiated to produce the waveshapes shown on the right. The resulting output can take the form of 14 different pulse widths above and under a "0" voltage step.
DSD data are inputted to the CXD8762 processor through a different digital interface, the data being already separated in a left and right serial format associated with a clock signal line by the DSD decoder chip, which also decrypts the various copy-protection of the SACD disc format. On fig. 1, the (7) on-chip DSD receiver called "SERIAL TEST" actually performs some signal processing on the DSD data to transform them in a format that is suitable to be used by the CXD9521 (or the built-in PWM converter inside the CXD8762). The process is clever and generates many additional opportunities. As seen just above about PCM data, the D/A converter has to be driven by 4 bits data. In order to use the same D/A converter for DSD, which have only two states, either "1" or "0", 1 bit data are translated by the digital input receiver in a 4 bits code. The 4 bits code is associated either to a pulse in the positive direction, or the same pulse but in the negative direction. As can be seen on fig. 3, the receiver (1) takes the 1 bit input DSD data and (2) translates all "1" into "0111" (+7 in decimal) and all "0" in to "1001" (-7 in decimal). This 4-bits translation lets the D/A converters produce in the analogue domain the desired pulses that correspond to each DSD sample. (3) One pair of D/A works differentially to produce a pulse length and its complement; (4) the other pair produces the inverted versions of said pulses. When differentially subtracted, the two pairs of signals respectively produce the final wave shapes shown in (5).
You may have noticed that the conversion of DSD data to analogue is not done according to the archetypal scheme of a string of pulses that are simply low-pass filtered. In the archetypal form seen everywhere else to explain how DSD works and is converted to analogue, the DSD bitstream produces an analogue waveform than only takes two alternating values, one positive, one negative (as can be seen in [1] on fig. 3). In the Sony system described in the patent, the analogue waveform have three states: one in the positive direction, one in the negative direction and one neutral state (at 0 V). According to Sony, this advantageously produces a systematic pattern of a leading edge and a trailing edge (or vice-versa) at each data clock cycle, whereas in the simplest canonical conversion process of 1 bit data, leading or trailing edge only happens when there is a transition from a 1 to a 0 or a 0 to a 1, but never when successive 1s or successive 0s follow each other. In other words, a signal-dependent switching distortion associated with 1-bit data becomes an evenly distributed switching noise that is decorrelated from the signal content and that is easier to be dealt with in the following analogue stages.
But that's not all.
As seven pairs of 4 bits codes are available, Sony has thought about using them as a digitally controlled analogue volume control. Because the mean analogue voltage at the output of the player is proportional to the width of the pulses after they are low-pass filtered, it is possible to use any of the seven available pairs of 4 bits codes to obtain seven different pulse widths, hence seven different gain settings, from 1/7 (lowest gain: pair of codes 0001 and 1111) to 7/7 (highest gain: pair of codes 0111 and 1001). Sony also envisaged to take the opportunity of the conversion of 1 bit DSD data in a 4 bits code to process DSD data digitally under the form of a very simple moving average low-pass filter to remove part of the shaped quantization noise. This very simple filter has three delay cells. Simple math shows that the decimal values that can be obtained by summing 4 successive samples having the value of unity (modulo the + or - sign) can be (if we begin from the bottom of the scale) -4, -3, -2, -1, 1, 2, 3, 4, i.e. eight different values, each corresponding to one of the fourteen available 4 bits code other than 0. That still leaves six other free 4 bits code than can be use to scale the output of the digital filter to form a 4-steps gain control. The output of the filter can thus be scaled to the values of -5, -4, -3, -2, 2, 3, 4, 5, or -6, -5, -4, -3, 3, 4, 5, 6, or finally -7, -6, -5, -4, 4, 5, 6, 7 to increase the output gain.
If we return to fig. 1, we can see that the CXD8762 has (8) two pins called PLMGAIN1 and PLMGAIN2. This pins are used with external DC voltages to hardware control the gain in the DSD receiver. The DC voltages can be either High or Low at each pin to form a 2 bits word, i.e. four different binary numbers, enough to realize a command bus to control the gain at the output of the digital low-pass filter just described. Therefore, we can confidently assume that the CXD8762 performs both the digital low-pass filtering of DSD data and the gain control described in the Sony patent.
There are many other interesting design aspects to discuss about this player, but I think I have already been to long. It was useful to share the main points to counterbalance so many inaccurate descriptions of the way DSD conversion and processing works in the early Sony SACD players.
Let's get to the review !
I will do my best to adhere to the following general framework, when it is possible with the various test discs available to me:
1. Dashboard @ about 1 kHz, 0 dBFS
2. Frequency response related measurements
3. Noise and distortion related measurements
4. Linearity tests
5. Special tests
6. Disc readability tests
Part II: Measurements of the Sony SCD-555ES as a CD Player
All measurements were taken with an Audio Precision System One+DSP SYS222A. Unless otherwise stated, the tests were performed with NTTY’s Test CD Version 7.2. The Audio Precision has always been given the 30 minutes preconditioning period mandated in its calibration procedure and the device under test the 5 minutes mandated by the AES-17 standard.
1. Dashboards
As said earlier, the Sony SCD-555ES gets five interpolation filters to choose from. Unless otherwise noted, all measurements below are performed with the "Standard" digital filter.
Keep in mind that the old Audio Precision System One has no dual analogue to digital converter per analyzer channel, hence cannot display a recombined representation of both the test signal and the distortion residual. Contrary to Amirm's reviews, you will only see the distortion residual with a “notch” on each side of the leftover of the test signal: this is where the analyzer band-reject filter has removed the test tone prior to the digitization and measurement of the residual signal.
The output level is in the vicinity of 2.15 V RMS. The levels of the two channels are matched to within 0.03 dB. The tests has also revealed a relatively close matching in performance metrics between the left and the right channels, but you nevertheless get 2 dashboards for the cost of one to appreciate that for yourself.
The mean SINAD of 96.5 dB is actually dominated by noise. You may have noticed the complete absence of visible power-supply related noise (mains frequency in Europe is at 50 Hz). The harmonic distortion spectra of the two channels are consistent with measurements made by a German laboratory in 2001 that I have reported elsewhere: the THD alone (noise excluded) was about -108.2 dB (0.00039%).
In find it interesting to show a dashboard of the most peculiar optional digital filter, numbered 2 :
The harmonic distortion pattern and overall THD+N do not change, but the output level is slightly lower and it can be seen that the notch due to the band-reject filter is less deep. Moreover, the THD+N bar-graph shows a wider standard deviation. I have seen similar effects with DSD sine signals. I think this is partly due to the DSD high frequency noise that lets the active notch filter enter slew rate limiting, but I can be wrong.
To comply with the practice set by NTTY in his exceptionally thorough CD player reviews, here are dashboards of the two channels configured the same but at -6 dBFS:
At -6 dBFS, the respective THD+N (inverse of SINAD) of the two channels are almost the same and are noise dominated. It is worth pointing out that the output level of the two channels matches perfectly, which is excellent.
2. Frequency response
Let's see how the five available interpolation filters of the Sony act on the player’s frequency response (only one channel shown). The two following measurements have been made with a glide tone at -15 dBFS from the Denon Audio Technical CD:
Filter 3, 4 and standard share more or less the same overall response, obviously imposed by the analogue output stage of the player (see also SACD measurements below for confirmation), save for the rate of attenuation of very high frequencies, whereas filter 1 and 2 roll the high frequencies more. Filter 4 has some ripple at the high frequencies. Another important effect is that filter 4 is down 0.25 dB in level compared to filter 1, 3 and standard, and filter 2 is down a good 0.125 dB more. This level differences alone will be audible during comparison.
Here is a high-resolution frequency response and inter-channel phase deviation plot of the player with the Standard digital filter:
The frequency response is tailored by the analogue output low-pass filter, with about -0.1 dB at 10 kHz and slightly less than -0.5 dB at 20 kHz. There is no significant deviation of phase response between the two channels.
Another way to look at the effect of the digital filters is to trace the wideband response with white noise up to ultrasonic frequencies (in this case 85 kHz).
The most interesting filter is number 3. Yes it attenuates more than the standard one at the top of the audio band, but its remains relatively sharp indeed and have a lower out of band noise floor. Sony says this is a single stage FIR filter.
The SCD-555ES decodes emphasized tones at -20 dBFS from the Stereophile Test CD2 at almost correct levels:
Decoding of emphasized data has also been assessed with tracks at -10 and -60 dBFS from the HiFi-News and Record Review Test CD II (HFN 015). The levels with this test tone are also correct.
Crosstalk evaluated with spot tones from the Denon Audio Technical CD is better in the right to left than the left to right direction, but is good in any case:
3. Noise and distortion measurements
Let's start with some Left/Right single point measurements about noise and distortion before proceeding with more graphs. All measurements are done with standard Audio Precision tests:
The signal to noise ratio is consistent with the dynamic range. That indicates that, contrary to an old practice followed by Sony until the SCD-555ES, the DAC does not mute its outputs when it is fed with a digital signal containing only zeros. The reason why the signal to noise ratio is greater than the dynamic range is because a digital signal containing only zeros does not exercise any quantization level above or below the 0 level, hence no quantization noise is generated at the output of the DAC.
The standard digital filter has no headroom to reproduce inter-sample overs, as do the 4 other optional filters. For the record, digital filter 2 degrades the intermodulation test figures rather dramatically (about -55 dB measured on both channels with DIN test signals at -10 and -1.68 dBFS), whereas the other optional filters remain more controlled.
To go further in the analysis, Audio Precision provides some standard tests to assess the noise performance of a CD player.
First test is to look at the FFT spectrum of an “Infinity Zero” signal from the Denon Audio Technical CD up to 80 kHz:
Please take into consideration than this measurement is free from any influence of the digital processing since the test signal contains only zeros. It only shows the analogue components of the noise. A variation of this test suggested by Audio Precision is to restrict the measurement bandwidth to the low frequencies and to lower the sample rate in order to narrow the bin frequency width. That way, it is possible to search for mains frequency interference and power supply related noise with great resolution:
As you can see, the FFT spectrum is very clean, with only 50 and 100 Hz tones present on both channels at very low levels, especially in the right channel.
The other tests use the analogue analyzer to plot 1/3 octave curves of the noise with the same "Infinity Zero" signal. Although this method has less frequency resolution than the FFT analysis (the higher the frequency, the wider the noise bandwidth that each point of the curve represents contrary to an FFT analysis where each “bin” has equal bandwidth; that is why the curve obtained with the analogue analyzer regularly increases towards higher frequencies), it takes advantage of the greater dynamic range and the wider frequency response of the Audio Precision System One’s analogue analyzer over its ADCs.
A good correlation can be seen on both the FFT and the wideband analogue noise spectrum.
The same test restricted to the audio bandwidth will be useful to compare to a special test that will be shown below:
Lastly, I add a test inspired by NTTY's practice to show an FFT spectrum up to 1 kHz when the player is reproducing a 1 kHz tone at 0 dBFS in search of power supply spurious noise. For this test, I set up an FFT at 8 ksps in order to improve bin frequency resolution to look after the output of the notch filter. Here I have to make a choice about which signal to use. I have chosen the 999.91 Hz sine with noise-shaped dither, which produces the lowest noise floor. The FFT is almost free of any power supply related spurious except at 250 Hz, but we can also see a smidgen of 100 Hz side-band at the left of the remainder of the test tone:
THD+N vs frequency has been assessed with spot tones at 0 dBFS without dither from the Pierre Vérany Digital Test CD. First within a 20 kHz bandwidth (i.e. the audio band):
There is no increase in distortion in the bass, but the THD+N is higher towards the high frequencies. The sharp downward slope of the curve above 10 kHz is due to the fact that the second harmonic at 10 kHz and above falls outside the pass-band of the test, hence there is no longer any harmonic to measure, only noise. The upward curve from about 15 kHz most probably indicates an increase in the noise (including quantization noise) or distortion due to imaging above the 22.05 kHz Nyquist frequency of CD folding back into the audio band when the player has to reproduce high frequencies at high level.
Second, here is the same test, but with a wider measurement bandwidth up to 80 kHz:
The increasing THD+N at higher frequencies is now more obvious.
THD+N in function of levels has also been assessed with the help the 999.91 Hz test signals from NTTY CD test disc. First with dithered tones (which gives a typical performance curve, because CD production is almost universally done with dither):
The flat curve up to a level of about -15 dB indicates that the measurement is dominated by the dither noise and that minimal excess distortion begins to appears at signal levels above -15 dBFS. It is possible to see a truer picture of the level of performance of the player with the same measurement but this time with the sine tones without dither:
Here, the two curves stay at or below -98 dB THD+N up to -15 dBFS level or so. That means that the Sony SCD-555ES is capable of 16 bits accuracy over almost a 45 dB wide dynamic range, and above -15 dBFS, little excess noise or distortion begin to appear.
Audio Precision provides a standard test to evaluate more qualitatively the SMPTE intermodulation distortion by looking at the digitization of the output of the analogue analyzer filter that removes the 60 Hz and 7 kHz tones to observe the actual distortion products:
And here is the same test with digital filter #2, which shows a much worsened distortion pattern:
The old 16 bits Burr Brown PCM78 ADCs of the Audio Precision System One probably limits the usefulness of FFTs of high level CCIF twin tone signals, because of the high dynamic of this signal relative to the noise floor. Here is an FFT of a -3.02 dBFS twin tones in a 23 kHz bandwidth:
Pairs of odd order distortion products (at 18 and 21 kHz, 17 and 22 kHz and 16 and 23 kHz) are visible, the highest pair being at about 90 dB under the level of each twin tones. The even order distortion at 1 kHz is some 100 dB under the level of each twin tones.
For the same reason already explained about the dynamic range of the old Audio Precision ADCs, I do not think an FFT of the multitone signal from NTTY's test CD can give much detailed information. Moreover, the maximum FFT length that the System One is capable of restricts the bin frequency resolution in the bass. Anyway, here is the FFT spectrum of such signal, but from 100 Hz only:
Lowering the sample rate of the FFT to 8 ksps and restricting the analysis to a lower 4 kHz bandwidth in order to observe only the lower part of the multitone signal gives a bit more resolution (not shown), but there is not much more to report.
4. Linearity Tests
With the kind help of NTTY, who has created a custom linearity test track with noise-shaped dithered spot tones down to -130 dBFS, I am able to assess the player's DAC deviation from linearity with the System One's software-implemented selective voltmeter:
I have used the same scale than the one Amirm uses for its own tests to ease comparison. The yellow curve (right channel) stops at -110 dBFS because the right channel of the Audio Precision is unable to settle to any reading under that level. Nevertheless, we quite certainly get an almost perfect linearity down to -120 dBFS on both channels as can be seen on the left channel, the worst case deviation down to that level being at most +/-0.1 dB. The deviation from linearity of the left channel at -130 dBFS assessed with NTTY's custom test signal is +1.3 dB, but that is not stable and does vary from one measurement to another, because the signal lies deep in the player's analogue noise. So, the exact level at which the player's output deviates significantly from linearity lies somewhere between -120 dBFS and -130 dBFS.
The above results have been obtained from measurements that I made only several dozens minutes after the player had been switched on. But on one instance, I made an FFT analysis of the -130 dBFS sine signals after half a day of operation and here is the outcome:
I don't know if the fact that I managed to randomly take a snapshot at the precise moment when the level of the two signals fell simultaneously almost exactly at the correct value was due to sheer luck or if there actually is an improvement of the linearity at extremely low levels due to thermal stabilization, but at least we can appreciate the cleanliness of the signal.
(To be continued in message #2)
This is a review and test of the Sony SCD-555ES SACD and CD player. This is an European and Japan model. It has been released at the end of 2000 and cost around 1,830 euros at the time (about $1,650 of 2000). It has no US equivalent.
Part I: Presentation
The Sony SCD-555ES is a stereo only SACD and CD player, without any digital output for DSD data, only for 16 bits @ 44.1 ksps PCM data. There is one set of RCA (unbalanced) output and a front phone output jack with volume control. The Sony offers 5 different digital interpolation filters for CD data to choose from.
This player belongs to the very first generation of SACD players from Sony. It is a direct hair of the SCD-1, the very first SACD player, and the almost identical SCD-777ES. In fact, the SCD-555ES is the last Sony player to have used Sony's own "Pulse Length Modulation" (PLM) digital to analogue conversion system that had been introduced in 1989 in the seminal DAS-R1a. NTTY has already reviewed two other CD players that use a Sony DAC chip of the same family: the CDP-X333ES and CDP-X559ES. Please go read the review of the former, where the Sony system is discussed in some length.
Those of you that are interested by technical design would be happy to know that the service manual of the SCD-555ES is readily available and, above all, that the theory of operation of the first generation of Sony SACD players has been very comprehensively described in a patent granted in 2001. As I have always found the ideas herein described very neat, I cannot resist to elaborate a bit on some core aspects of the design. The main goal of Sony was to get a conversion system that has as much commonalities as possible between CD and SACD replay circuits. Actually, it is not an exaggeration to say that Sony's first SACD players were actually regular Sony CD players that had somehow been hacked to also play DSD programs with minimal changes.
The heart of the player is the Sony CXD8762 digital signal processor (figure 1):
This chip takes the serial interleaved left/right PCM input (which can goes up to 24 bits depth, although only 16 bits are used in this player, no re-quantization being made inside the DSP), it de-interlaces the left and right data-stream in a parallel format and performs 8 times digital interpolation (1) to increase the sample rate to 352.8 kHz. Sony advertises some exotic optional "slow roll-off" filters that are user selectable. Apart from the usual high frequency roll-off associated with that kind of filters, the most interesting aspect of at least some of them is that the interpolation is not done in successive cascaded FIR filters, but in a single-stage FIR filter thanks to the processing power available on-chip. A standard interpolation filter based on a 3-stages FIR filter is nevertheless provided. After this first interpolation, a further 8 times oversampling is performed by a (2) linear interpolator (i.e. the computed additional samples follow a straight line from one original sample to the next original sample). A this point, the sample rate becomes 2.8224 MHz. Not coincidentally, this sample rate is the same as DSD. Then, the digital data are re-quantized to 15 levels coded with 4 bits in a 2-complement forms by the (3) noise-shaper (probably third order, as many other Sony chips are). The (4) PWM block is an actual digital to analogue converter which gives the option to use the chip as an integrated digital processor and DAC, but this converter stage is not used in the Sony SCD-555ES. The on-chip converter is by-passed by (5) solid-state logic switches and the 4 bits data are put out in a parallel format at the (6) stereo outputs of the chip. This switches also select the data coming from the DSD signal path. Incidentally, when this chip has been used as an integrated digital processor and DAC in some players, Sony gave it a different part number, CXD9556, but the chips bearing both designations are actually identical.
In the SCD-555ES, the D/A conversion is done in a following chip, the CXD9521. The conversion is a double differential method, i.e. the digital data are converted by four D/A working simultaneously to produce complement and inverted signals that are successively differentially subtracted in the analogue domain. The output waveform of one pair of D/A stage is as shown on fig. 2 (the other pair shapes the inverse of the shown waveforms). T corresponds to the period of the master converter clock, which is 45.1584 MHz, 16 times the rate of the noise-shaped data.
The (1) first D/A of the differential pair converts the 4 bits input data in pulse length proportional to the input data, the (2) second D/A of the pair produces a pulse length that is the complement of the the first one and the two pulses are (3) differentiated to produce the waveshapes shown on the right. The resulting output can take the form of 14 different pulse widths above and under a "0" voltage step.
DSD data are inputted to the CXD8762 processor through a different digital interface, the data being already separated in a left and right serial format associated with a clock signal line by the DSD decoder chip, which also decrypts the various copy-protection of the SACD disc format. On fig. 1, the (7) on-chip DSD receiver called "SERIAL TEST" actually performs some signal processing on the DSD data to transform them in a format that is suitable to be used by the CXD9521 (or the built-in PWM converter inside the CXD8762). The process is clever and generates many additional opportunities. As seen just above about PCM data, the D/A converter has to be driven by 4 bits data. In order to use the same D/A converter for DSD, which have only two states, either "1" or "0", 1 bit data are translated by the digital input receiver in a 4 bits code. The 4 bits code is associated either to a pulse in the positive direction, or the same pulse but in the negative direction. As can be seen on fig. 3, the receiver (1) takes the 1 bit input DSD data and (2) translates all "1" into "0111" (+7 in decimal) and all "0" in to "1001" (-7 in decimal). This 4-bits translation lets the D/A converters produce in the analogue domain the desired pulses that correspond to each DSD sample. (3) One pair of D/A works differentially to produce a pulse length and its complement; (4) the other pair produces the inverted versions of said pulses. When differentially subtracted, the two pairs of signals respectively produce the final wave shapes shown in (5).
You may have noticed that the conversion of DSD data to analogue is not done according to the archetypal scheme of a string of pulses that are simply low-pass filtered. In the archetypal form seen everywhere else to explain how DSD works and is converted to analogue, the DSD bitstream produces an analogue waveform than only takes two alternating values, one positive, one negative (as can be seen in [1] on fig. 3). In the Sony system described in the patent, the analogue waveform have three states: one in the positive direction, one in the negative direction and one neutral state (at 0 V). According to Sony, this advantageously produces a systematic pattern of a leading edge and a trailing edge (or vice-versa) at each data clock cycle, whereas in the simplest canonical conversion process of 1 bit data, leading or trailing edge only happens when there is a transition from a 1 to a 0 or a 0 to a 1, but never when successive 1s or successive 0s follow each other. In other words, a signal-dependent switching distortion associated with 1-bit data becomes an evenly distributed switching noise that is decorrelated from the signal content and that is easier to be dealt with in the following analogue stages.
But that's not all.
As seven pairs of 4 bits codes are available, Sony has thought about using them as a digitally controlled analogue volume control. Because the mean analogue voltage at the output of the player is proportional to the width of the pulses after they are low-pass filtered, it is possible to use any of the seven available pairs of 4 bits codes to obtain seven different pulse widths, hence seven different gain settings, from 1/7 (lowest gain: pair of codes 0001 and 1111) to 7/7 (highest gain: pair of codes 0111 and 1001). Sony also envisaged to take the opportunity of the conversion of 1 bit DSD data in a 4 bits code to process DSD data digitally under the form of a very simple moving average low-pass filter to remove part of the shaped quantization noise. This very simple filter has three delay cells. Simple math shows that the decimal values that can be obtained by summing 4 successive samples having the value of unity (modulo the + or - sign) can be (if we begin from the bottom of the scale) -4, -3, -2, -1, 1, 2, 3, 4, i.e. eight different values, each corresponding to one of the fourteen available 4 bits code other than 0. That still leaves six other free 4 bits code than can be use to scale the output of the digital filter to form a 4-steps gain control. The output of the filter can thus be scaled to the values of -5, -4, -3, -2, 2, 3, 4, 5, or -6, -5, -4, -3, 3, 4, 5, 6, or finally -7, -6, -5, -4, 4, 5, 6, 7 to increase the output gain.
If we return to fig. 1, we can see that the CXD8762 has (8) two pins called PLMGAIN1 and PLMGAIN2. This pins are used with external DC voltages to hardware control the gain in the DSD receiver. The DC voltages can be either High or Low at each pin to form a 2 bits word, i.e. four different binary numbers, enough to realize a command bus to control the gain at the output of the digital low-pass filter just described. Therefore, we can confidently assume that the CXD8762 performs both the digital low-pass filtering of DSD data and the gain control described in the Sony patent.
There are many other interesting design aspects to discuss about this player, but I think I have already been to long. It was useful to share the main points to counterbalance so many inaccurate descriptions of the way DSD conversion and processing works in the early Sony SACD players.
Let's get to the review !
I will do my best to adhere to the following general framework, when it is possible with the various test discs available to me:
1. Dashboard @ about 1 kHz, 0 dBFS
2. Frequency response related measurements
3. Noise and distortion related measurements
4. Linearity tests
5. Special tests
6. Disc readability tests
Part II: Measurements of the Sony SCD-555ES as a CD Player
All measurements were taken with an Audio Precision System One+DSP SYS222A. Unless otherwise stated, the tests were performed with NTTY’s Test CD Version 7.2. The Audio Precision has always been given the 30 minutes preconditioning period mandated in its calibration procedure and the device under test the 5 minutes mandated by the AES-17 standard.
1. Dashboards
As said earlier, the Sony SCD-555ES gets five interpolation filters to choose from. Unless otherwise noted, all measurements below are performed with the "Standard" digital filter.
Keep in mind that the old Audio Precision System One has no dual analogue to digital converter per analyzer channel, hence cannot display a recombined representation of both the test signal and the distortion residual. Contrary to Amirm's reviews, you will only see the distortion residual with a “notch” on each side of the leftover of the test signal: this is where the analyzer band-reject filter has removed the test tone prior to the digitization and measurement of the residual signal.
The output level is in the vicinity of 2.15 V RMS. The levels of the two channels are matched to within 0.03 dB. The tests has also revealed a relatively close matching in performance metrics between the left and the right channels, but you nevertheless get 2 dashboards for the cost of one to appreciate that for yourself.
The mean SINAD of 96.5 dB is actually dominated by noise. You may have noticed the complete absence of visible power-supply related noise (mains frequency in Europe is at 50 Hz). The harmonic distortion spectra of the two channels are consistent with measurements made by a German laboratory in 2001 that I have reported elsewhere: the THD alone (noise excluded) was about -108.2 dB (0.00039%).
In find it interesting to show a dashboard of the most peculiar optional digital filter, numbered 2 :
The harmonic distortion pattern and overall THD+N do not change, but the output level is slightly lower and it can be seen that the notch due to the band-reject filter is less deep. Moreover, the THD+N bar-graph shows a wider standard deviation. I have seen similar effects with DSD sine signals. I think this is partly due to the DSD high frequency noise that lets the active notch filter enter slew rate limiting, but I can be wrong.
To comply with the practice set by NTTY in his exceptionally thorough CD player reviews, here are dashboards of the two channels configured the same but at -6 dBFS:
At -6 dBFS, the respective THD+N (inverse of SINAD) of the two channels are almost the same and are noise dominated. It is worth pointing out that the output level of the two channels matches perfectly, which is excellent.
2. Frequency response
Let's see how the five available interpolation filters of the Sony act on the player’s frequency response (only one channel shown). The two following measurements have been made with a glide tone at -15 dBFS from the Denon Audio Technical CD:
Filter 3, 4 and standard share more or less the same overall response, obviously imposed by the analogue output stage of the player (see also SACD measurements below for confirmation), save for the rate of attenuation of very high frequencies, whereas filter 1 and 2 roll the high frequencies more. Filter 4 has some ripple at the high frequencies. Another important effect is that filter 4 is down 0.25 dB in level compared to filter 1, 3 and standard, and filter 2 is down a good 0.125 dB more. This level differences alone will be audible during comparison.
Here is a high-resolution frequency response and inter-channel phase deviation plot of the player with the Standard digital filter:
The frequency response is tailored by the analogue output low-pass filter, with about -0.1 dB at 10 kHz and slightly less than -0.5 dB at 20 kHz. There is no significant deviation of phase response between the two channels.
Another way to look at the effect of the digital filters is to trace the wideband response with white noise up to ultrasonic frequencies (in this case 85 kHz).
The most interesting filter is number 3. Yes it attenuates more than the standard one at the top of the audio band, but its remains relatively sharp indeed and have a lower out of band noise floor. Sony says this is a single stage FIR filter.
The SCD-555ES decodes emphasized tones at -20 dBFS from the Stereophile Test CD2 at almost correct levels:
Decoding of emphasized data has also been assessed with tracks at -10 and -60 dBFS from the HiFi-News and Record Review Test CD II (HFN 015). The levels with this test tone are also correct.
Crosstalk evaluated with spot tones from the Denon Audio Technical CD is better in the right to left than the left to right direction, but is good in any case:
3. Noise and distortion measurements
Let's start with some Left/Right single point measurements about noise and distortion before proceeding with more graphs. All measurements are done with standard Audio Precision tests:
The signal to noise ratio is consistent with the dynamic range. That indicates that, contrary to an old practice followed by Sony until the SCD-555ES, the DAC does not mute its outputs when it is fed with a digital signal containing only zeros. The reason why the signal to noise ratio is greater than the dynamic range is because a digital signal containing only zeros does not exercise any quantization level above or below the 0 level, hence no quantization noise is generated at the output of the DAC.
The standard digital filter has no headroom to reproduce inter-sample overs, as do the 4 other optional filters. For the record, digital filter 2 degrades the intermodulation test figures rather dramatically (about -55 dB measured on both channels with DIN test signals at -10 and -1.68 dBFS), whereas the other optional filters remain more controlled.
To go further in the analysis, Audio Precision provides some standard tests to assess the noise performance of a CD player.
First test is to look at the FFT spectrum of an “Infinity Zero” signal from the Denon Audio Technical CD up to 80 kHz:
Please take into consideration than this measurement is free from any influence of the digital processing since the test signal contains only zeros. It only shows the analogue components of the noise. A variation of this test suggested by Audio Precision is to restrict the measurement bandwidth to the low frequencies and to lower the sample rate in order to narrow the bin frequency width. That way, it is possible to search for mains frequency interference and power supply related noise with great resolution:
As you can see, the FFT spectrum is very clean, with only 50 and 100 Hz tones present on both channels at very low levels, especially in the right channel.
The other tests use the analogue analyzer to plot 1/3 octave curves of the noise with the same "Infinity Zero" signal. Although this method has less frequency resolution than the FFT analysis (the higher the frequency, the wider the noise bandwidth that each point of the curve represents contrary to an FFT analysis where each “bin” has equal bandwidth; that is why the curve obtained with the analogue analyzer regularly increases towards higher frequencies), it takes advantage of the greater dynamic range and the wider frequency response of the Audio Precision System One’s analogue analyzer over its ADCs.
A good correlation can be seen on both the FFT and the wideband analogue noise spectrum.
The same test restricted to the audio bandwidth will be useful to compare to a special test that will be shown below:
Lastly, I add a test inspired by NTTY's practice to show an FFT spectrum up to 1 kHz when the player is reproducing a 1 kHz tone at 0 dBFS in search of power supply spurious noise. For this test, I set up an FFT at 8 ksps in order to improve bin frequency resolution to look after the output of the notch filter. Here I have to make a choice about which signal to use. I have chosen the 999.91 Hz sine with noise-shaped dither, which produces the lowest noise floor. The FFT is almost free of any power supply related spurious except at 250 Hz, but we can also see a smidgen of 100 Hz side-band at the left of the remainder of the test tone:
THD+N vs frequency has been assessed with spot tones at 0 dBFS without dither from the Pierre Vérany Digital Test CD. First within a 20 kHz bandwidth (i.e. the audio band):
There is no increase in distortion in the bass, but the THD+N is higher towards the high frequencies. The sharp downward slope of the curve above 10 kHz is due to the fact that the second harmonic at 10 kHz and above falls outside the pass-band of the test, hence there is no longer any harmonic to measure, only noise. The upward curve from about 15 kHz most probably indicates an increase in the noise (including quantization noise) or distortion due to imaging above the 22.05 kHz Nyquist frequency of CD folding back into the audio band when the player has to reproduce high frequencies at high level.
Second, here is the same test, but with a wider measurement bandwidth up to 80 kHz:
The increasing THD+N at higher frequencies is now more obvious.
THD+N in function of levels has also been assessed with the help the 999.91 Hz test signals from NTTY CD test disc. First with dithered tones (which gives a typical performance curve, because CD production is almost universally done with dither):
The flat curve up to a level of about -15 dB indicates that the measurement is dominated by the dither noise and that minimal excess distortion begins to appears at signal levels above -15 dBFS. It is possible to see a truer picture of the level of performance of the player with the same measurement but this time with the sine tones without dither:
Here, the two curves stay at or below -98 dB THD+N up to -15 dBFS level or so. That means that the Sony SCD-555ES is capable of 16 bits accuracy over almost a 45 dB wide dynamic range, and above -15 dBFS, little excess noise or distortion begin to appear.
Audio Precision provides a standard test to evaluate more qualitatively the SMPTE intermodulation distortion by looking at the digitization of the output of the analogue analyzer filter that removes the 60 Hz and 7 kHz tones to observe the actual distortion products:
And here is the same test with digital filter #2, which shows a much worsened distortion pattern:
The old 16 bits Burr Brown PCM78 ADCs of the Audio Precision System One probably limits the usefulness of FFTs of high level CCIF twin tone signals, because of the high dynamic of this signal relative to the noise floor. Here is an FFT of a -3.02 dBFS twin tones in a 23 kHz bandwidth:
Pairs of odd order distortion products (at 18 and 21 kHz, 17 and 22 kHz and 16 and 23 kHz) are visible, the highest pair being at about 90 dB under the level of each twin tones. The even order distortion at 1 kHz is some 100 dB under the level of each twin tones.
For the same reason already explained about the dynamic range of the old Audio Precision ADCs, I do not think an FFT of the multitone signal from NTTY's test CD can give much detailed information. Moreover, the maximum FFT length that the System One is capable of restricts the bin frequency resolution in the bass. Anyway, here is the FFT spectrum of such signal, but from 100 Hz only:
Lowering the sample rate of the FFT to 8 ksps and restricting the analysis to a lower 4 kHz bandwidth in order to observe only the lower part of the multitone signal gives a bit more resolution (not shown), but there is not much more to report.
4. Linearity Tests
With the kind help of NTTY, who has created a custom linearity test track with noise-shaped dithered spot tones down to -130 dBFS, I am able to assess the player's DAC deviation from linearity with the System One's software-implemented selective voltmeter:
I have used the same scale than the one Amirm uses for its own tests to ease comparison. The yellow curve (right channel) stops at -110 dBFS because the right channel of the Audio Precision is unable to settle to any reading under that level. Nevertheless, we quite certainly get an almost perfect linearity down to -120 dBFS on both channels as can be seen on the left channel, the worst case deviation down to that level being at most +/-0.1 dB. The deviation from linearity of the left channel at -130 dBFS assessed with NTTY's custom test signal is +1.3 dB, but that is not stable and does vary from one measurement to another, because the signal lies deep in the player's analogue noise. So, the exact level at which the player's output deviates significantly from linearity lies somewhere between -120 dBFS and -130 dBFS.
The above results have been obtained from measurements that I made only several dozens minutes after the player had been switched on. But on one instance, I made an FFT analysis of the -130 dBFS sine signals after half a day of operation and here is the outcome:
I don't know if the fact that I managed to randomly take a snapshot at the precise moment when the level of the two signals fell simultaneously almost exactly at the correct value was due to sheer luck or if there actually is an improvement of the linearity at extremely low levels due to thermal stabilization, but at least we can appreciate the cleanliness of the signal.
(To be continued in message #2)
Last edited:
The 3DC trace is neat and is an additional indication of low noise, probably limited by the System One’s resolution. I don’t know if it has a scaler, but if not try to increase the gain at the input of the analyzer, it could help lowering the noise of the System One and get a better view. As is it is already near perfect, so…
