Integrated Amplifiers

Harman Kardon HK 990 Stereo Integrated Amplifier with Digital Room Correction and Dual Subwoofer Bass Management – Part II

ARTICLE INDEX

HK 990 Digital Signal Selector, Clock and Data Recovery, Jitter Reduction, and Digital Reconstruction Filtering

The block diagram of figure 3 above shows the digital input signal flows. The electronics are on the right-most daughter board. Its placement is defined by the optical and coax digital input jacks on the rear panel. The clock and data from the Sony/Philips Digital Interconnect Format (SPDIF) digital inputs are recovered by a TI SRC4392. The chip also functions as the input selector for the digital signals. In an AVR, digital audio data from the HDMI receivers can be selected in this system block. The HDMI receivers are usually on the video board, which is typically at the top of the audio boards in an AVR, which explains the placement of the HDMI plugs atop the rear panel in most AVRs.

The "Phase-Locked Loop" (PLL) is a basic circuit block for the clock and data recovery. Although the SPDIF signal is a serial bit stream, it is coded so the continuous clock signal and the signal that contains the LPCM data can be recovered and separated (the clock and data are shown together in the single thick red wire.). A PLL is on the die of the TI SRC4392. The mechanics of a PLL, its application in a clock and data recovery system, and its ability to reduce jitter, are best left for another discussion. Many texts have been written on Phase-Locked Loop design in both the analog and digital domains. I will try to summarize the essentials in the Primer at some point.

A two-channel (for stereo) asynchronous sample-rate converter (ASRC) resides on the TI SRC4392 to resolve an issue that the recovered clock from the SPDIF signal is not in sync with the clock that drives the DACs. The ASRC also provides digital filtering to remove signals above 20kHz (for the case of a CD) that occur during the sampling process. I discuss ASRCs in more detail below.

A. Maintaining High-resolution Data Through the Digital Signal Selector Front End

1) Bit depth

Figure 4 above shows the audiophile currently has a wealth of high resolution downloadable formats. Whether a modern unit with SPDIF inputs accepts and recovers 24bit incoming data from high-resolution material (DVD-A disc; WAV file; FLAC file) is a less than transparent issue, but one that is significant. Perhaps, more important, is the truncation of the signal as it makes its way from the SPDIF receiver to the DACs. To determine if the 24bits are preserved, the amplitude of a dithered signal is examined as it is reduced from -90dB below full scale (this is called the amplitude linearity test). The in-band signal-to-noise ratio (relative to full scale) is also monitored during the test. The test should occur when the complex DAC process options, such as room correction and bass management, are selected.

For all cases, the signal-to-noise ratios and amplitude linearity are limited by the DAC in the digital path if the 24bit word makes it to the DAC. The product sheet on the HK 990 is unclear about transmission of the 24bit words, but Harman Labs sent me data that indicating the 24bit signal was passed with the DSP in the signal path. The measurements covered the complete analog signal path to preamp out, including the volume control. The HK 990 measurements from Harman labs showed the equivalent signal-to-noise ratio was 19 equivalent bits. A 21bit signal had a -0.7 error relative to its expected amplitude. At 22bit word (-132dB from full scale) had a -1.3dB error. These results are consistent with the performance of the parts used in the HK 990 which will be discussed below.

The signal-to-noise (SNR) ratio of a signal recorded with a 16-bit quantization (98dB) cannot be lowered by signal processing on the received data as some high end companies imply. The number of bits that represent the signal grows as the signal progresses through the digital filters and other digital signal processes. The additional bits occur as mathematical operations are performed on the data. Failure to keep these additional bits may cause distortion and frequency response errors.

Some CDs are recorded with shaped quantization noise.  The noise is shaped by removing it from areas of the spectrum that the ear is most sensitive to it. Since the total noise power over the full band must be constant, more noise power is placed at areas of spectrum where the ear is less sensitive to noise. Quantization noise can be shaped only during the recording process. Noise shaping must be done before the data is pressed on the CD. If a noise shaper was not used the process cannot replicated from data coming of the CD.

The noise floor of fast Fourier transform (FFT) developed spectrums shown in audio equipment reports is not the signal-to-noise ratio. The noise floor is the noise over a small frequency band. The size of the band depends on parameters of the FFT.  This is shown in Figure 5 below.

2) Sampling Rate

Both the bit depth from the SPDIF digital input and the sampling rate of the high-resolution input must be preserved. The HK 990 accepts inputs with sampling rates up to 192kHz. Like the bit depth, the issue is not whether the SPDIF receiver locks-up to the incoming signal, but the quality of the signal received by the DAC. Activating a function such as room correction may result in sub-sampling (96kHz sampling reduced to 48kHz) in some products.

To identify sub-sampling, the spectra at the preamp output are monitored as a high-resolution file is connected to the input. Sub-sampling has occurred when the spectra above 20KHz are missing.

Again the product information for the HK 990 does not provide the critical information but data from Harman Labs, showed the HK 990 passes the full spectra (45kHz) of a 96kHz input, but this did not increase for a 192kHz input. It is hard to believe anybody could discern the difference.

The SPDIF output mutes on any SACD player when an SACD is inserted. Instead, one must use the analog output on the SACD player and run it to the HK 990 analog inputs. This is not an issue if you use the HK 990 in the digital bypass mode. If you want to enable room correction or bass management the HK 990 must first convert the analog signal from the SACD player back to digital.

An AVR attached to a Universal DVD player by HDMI will transfer a SACD digital data (DSD or transcoded to LPCM) since HDMI encrypts digital data. But dealing with video setup screens on both the AVR and Universal DVD player to enable the link can be complicated, and stereo-centric audiophiles will find this task tedious beyond belief. You need a TV in the room to set things up. The TV can be removed after setup, but may be called back should either the universal DVD player and/or AVR forget its settings as a result of a glitch like a power outage.

The fact that the operability of the HK 990 is not display-dependent is what makes it so wonderful. If you are going to use the room correction or bass management of the HK 990 the best thing to do is use a good CD player and not an SACD player. The CD player will read the CD layer of the SACD and transfer that over SPDIF. The redundant DAC –ADC conversions are eliminated. Given the inability to get high resolution data off an SACD using SPDIF it is clear that high resolution downloads are the way to go in stereo. SACD is important only to the multichannel listener.

B. The Performance and Viability of Proprietary Digital Interfaces

Harman's proprietary digital interface to its CD players, called the HRS link, provides a return path for the HK 990 system clock to the Harman HD 990 CD player (not shown in the digital input signal flows block diagram above). Tyler Stripko reviewed the HD 990. Figure 6 below shows the special HRS Link multi-wire jack.

As discussed above, digital devices with an SPDIF transmitter have clocks that are asynchronous to the main crystal oscillator on the HK 990. It is impossible to match the crystal oscillator frequency on the device transmitting the SPDIF signal to the crystal oscillator on the HK 990. The HRS link resolves the problem for the HD 990 CD player by forcing the HD 990 CD player to slave the HK 990 clock.

The crystal clock oscillator in the HD990 CD player is disabled when the HRS system is operative. Jitter on the HRS return path is not an issue because the clock that drives the DACs is not recovered from it. The DACs are clocked with the crystal oscillator on the DSP board, which is very close to the DAC.

These types of proprietary links emerged when SPDIF first appeared. Sony's first solution was an extra SPDIF cable running the clock back to the CD player. Unfortunately, comparable systems of different manufacturers are not fungible. Sony no longer supports the extra SPDIF cable for clock return. Marantz has a BNC input on some of its high-end SACD players marked External Master Clock.

The latest Sony proprietary-link system is called H.A.T.S. for its HDMI connections. The HDMI 1.3 CEC return line controls the data transmission rate to the universal player based on the clock in the AVR. You need a Sony AVR and Sony Blu-ray player that supports H.A.T.S. Pioneer has a nearly identical, but not inter-operative, system called PQLS. Both H.A.T.S. and PQLS are derived from the Audio Rate Control (ARC), which is a new function in version 1.3 of the HDMI spec. Games of deviating from the HDMI spec to create proprietary modes of operation caused big box stores to insist on absolute HDMI interoperability between all suppliers. This appears to have wiped the Audio Rate Control (ARC) off the latest generation of products.

Wedding yourself to these proprietary links can create long term problems. Disk players often require expensive repairs after warranty up to the point where unit replacement is more economical. One is unlikely to find a new unit that supports the old proprietary link. Also consider the circumstances of audiophiles with high-end universal DVD players that had proprietary links when Blu-ray emerged.

Also note that, most units that store and stream high-resolution files only have standard SPDIF outputs. The takeaway is to be certain the equipment you purchase has electronics that insurer the SNR and THD of the analog output is not degraded by impairments of the equipment sourcing the SPDIF signal or impairments from long runs of the SPDIF cable.

C. Working with Asynchronous Clocks without a Proprietary Link in Systems with DSP Processors

As discussed above the clock frequency of a unit with the SPDIF receiver differs from the clock in the unit with the SPDIF receiver. One solution to the problem is the proprietary link. A second alternative is to clock all electronics in the unit with the recovered clock from the source device. The recovered clock must have very low jitter when it clocks the DAC or SNR and THD will be degraded. Methods are available to achieve this. Note direct measurements on the clock sent to the DAC in the presence of jitter at the input can guide the designer in optimizing the system. It is possible for a reviewer to open up the box and make these measurements but I strongly dislike this. Any effect of clock jitter should be observable at the analog outputs of the unit in the form of increased noise or new distortion components.

In units like the HK 990, and most AVRs, the digital electronics in the unit (often multiple DSPs) are clocked by a local crystal oscillator. A digital circuit block, called an Asynchronous Sample Rate Converter (ASRC), is used. The recovered clock and LPCM data (one channel) from the SPDIF receiver enters the ASRC. In addition, the local clock oscillator on HK 990 or AVR, enters the ASRC.

The ASRC processes the incoming LPCM data and the LPCM data leaves the chip is not the same as what came in.  To understand this, let us look at a conceptual ASRC. An ideal DAC is clocked with one oscillator. The DAC, after reconstruction filtering, drives an ideal ADC clocked by a different oscillator. This conceptual ASRC is illustrated as a block diagram in Figure 7 below.

Note that the ASRC does not produce a clock it just provides a mechanism to interface the LPCM signal between to clocks that already exist. Clearly, the LPCM data entering the ideal DAC is not the same as what is exiting the ideal ADC if the clocks are different frequencies.

The ASRC can take in a clock with jitter on the LPCM input side and a jitter-free clock on the LPCM output side. Does this not imply all the jitter has been rejected. No! The effects of the jitter (increased SNR and distortion above what is measured with jitter-free clocks) may be present on the LPCM data exiting the ASRC. The worst-case situation is the mixed-signal example shown above. The DAC rejects none of the jitter on the clock driving it All the distortion and noise resulting from the jitter is now part of the LPCM data leaving the ADC.

ASRCs, in practice, are fully digital and have no internal DAC or ADCs. A fully digital ASRC interpolates (up samples) the LPCM input data to a very high sampling rate, which is then re-sampled at the output clock's rate. The system just described is impractical to integrate. Digital designers have developed systems that can emulate the functionality in less silicon.

Artifacts from the design limitations for a practical all-digital ASRC are small frequency response variations and distortion components in the LPCM data at the output. These arise in the process of the sample-rate conversion (no jitter on the clocks). The distortion components may not be harmonically related to the incoming tone encoded in the LPCM. The distortion and frequency response deviations are called out in the ASRCs data sheet for standard products like the TI SRC4392 used in the HK 990.

Distortion and frequency response variations can be measured on LPCM coming out of the ASRC; indeed, modern analog test equipment first converts the signal to LPCM before analysis. Different graphs are supplied for common sample rate conversions in audio products.

A practical fully-digital chip ASRC will reject some jitter. A section of the digital circuitry estimates the frequency ratio of the two clocks. The rate at which this estimate can change is limited. Robert Adams, who created the first ASRC for an audio application, points out the estimate is "computed using thousands of past input and output sample clock events and is therefore immune to small perturbations in the arrival time of any clock edge."

The ratio estimate is less affected the faster the clock edges arrival time changes. Some ASRC data sheets offer a graph that demonstrates how fast the clock edge arrival times must vary for the ratio estimator to become insensitive to it. This graph cannot replace graphs showing SNR and distortion increase with different types of jitter present on one of the clocks.  I have not seen an ARSC data sheet with these measurements.

Without a SNR and distortion measurements at the output of the ARSC in the presence of clock jitter at one of the clocks, a relative evaluation of two ASRC designs is not possible, and I cannot identify from the current data sheets which ARSC is best with respect to jitter rejection. Valid comparisons between standard products are obscured even further because the designer customizes the ASRCs performance by selecting internal options.

Murkier still, there are no standard tests that simulate the statistics of jitter that might be on the recovered SPDIF clock presented to the ASRC. First the test clock must simulate the statistics of jitter of the oscillator that is part of the PLL in the SPDIF receiver. At the SDPIF receiver recovered clock jitter statistics can change dependent on the LPCM data being transmitted (recall the clock and data are encoded on the single SPDIF cable) This is called data dependent jitter. Some LPCM test patterns have been proposed that are said produce worst case data dependent jitter but no consensus exists on which patterns are worst case. Recovered clock jitter will increase as the SPDIF signal passes through longer lengths of cable. Again this jitter may have different statistics.

For those who want to know more about jitter, how it affects DACs, and how an ASRC operates, please consider an article written by Robert Adams. It is published on pages 11 -22 of Issue 19 (Spring 1994) of The Audio Critic. The complete issue is available as a free PDF download on The Audio Critic website.

ASRCs may be found as part of a DSP chip (eight are in the Analog Devices ADSP-21469 designed for use in AVRs). An ASRC can be executed as software and implemented as an independent DSP or a DSP that performs multiple processing functions.

The service manual provides clues about the performance of the analog and data converter stages that can predict the performance when device measurements are made at the units rear panel as I will demonstrate in Part 3 of the review. In contrast, the parts used in the product provide no clue to the jitter rejection. Only special measurements at analog outputs will expose how much the jitter is attenuated. Without standardized tests I have chosen not to report on SNR and distortion impairments at the analog output as a result of jitter on the SPDIF input.

D. Digital Filtering of the LPCM Signal

An ASRC may provide the LPCM data at its output at a higher sampling rate than what came in. For example, LPCM data from a CD player arrives at 44.1kHz input, but could leave at 192kHz.

Artifacts from the sampling process, called images (shown in figure 8 below), are removed during the process of increasing the sampling rate.

For a band-limited 20kHz analog signal sampled at a rate of 44.1kHz samples/sec (fs) the spectra appear around 0Hz (desired signal) as well as from 24.1kHz to 64.1kHz and then repeat around multiples of 44.1kHz. The repeating spectra are the stop-band images from the sampling process. For a 20kHz signal (fB), the right side of the first image (often called the folded frequency) is close-in at 22.1kHz (44.1kHz – 20kHz). The suppression of the stop-band images allows the sampled signal to better represent an analog signal sampled at 44.1kHz. Information lost from band limiting of the signal before the sampling process during recording cannot be restored although some manufactures try to imply this. For high resolution files fs will be between 88.2 – 192kHz. The maximum frequency that can be recorded (fB) can also be extended since the right side of the first image will be at a higher frequency.

How well the ASRC removes the images is again dependent on its design. The process of attenuating the stop-band images of a sampled signal and increasing the sampling rate digitally often takes on interesting names from each company. For the HK 990, it is named the fourth-generation of the Real Time Linear Smoothing system (RLS IV). To repeat although the sample rate has been increased, no information about the sampled signal above 20kHz (for the case of a CD player) can be recreated. Instead, what has been accomplished is the removal of the folded tones in the digital domain to obviate removal later by complex analog filters.

In the absence of an ASRC a standard digital filter up-samples the incoming signal at an integer rate of 2, 4 or 8 and removes the stop band images. The filter may by as implemented as DSP software or it may be done in digital hardware. Like the ASRC, the filter leaves small frequency response variations and distortion products (folded tones not completely removed, for example). The design of the filter determines how well it performs in the frequency and time domains. Sometimes a standard digital filter may be placed after an ASRC to reduce the stop band images further.

Often the digital filter used is in the DAC chip itself, not an external block. The internal DAC filter is bypassed by the HK 990 in favor of the ASRC and perhaps additional filtering in the DSP block. The performance of the filter in the integrated circuit DAC may vary. Filters with poorer performance have more amplitude variation across the audible band, although this is normally dwarfed by the variation from the analog circuits. The other difference in filter performance is the attenuation of the folded spectra.

In general, the performance of the digital filter in the integrated DAC correlates with the THD and SNR performance. For superior performance in the mixed signal domain, more silicon area is made available for the digital filter because the improvement in analog noise floors can expose problems with the simpler digital filters.

Designers may design custom ASRCs and up-sampling filters to gain a performance advantage over generic products. Typically, the designer will do this in software using a standalone DSP. Some designers develop the filters to achieve performance objectives in the time and/or frequency domains. The custom DSP implementation is the most expedient way in which a designer can distinguish the performance of his product from the competition. This assumes a DAC has been selected with state of the art performance. In addition analog components connected to the DAC must be selected so they do not degrade the performance of the DAC. The Harman HD 990 CD player used proprietary software for the ASRC and up-sampling filter. Harman calls the system RLS III. An Analog Devices Blackfin DSP (Package photo shown in Figure 9 above, ® Analog Devices) was selected to execute the custom DSP code.

Like jitter, folded tone suppression must be measured. Examining the unit provides no clues, especially if the filter has custom crafted code for a DSP. Unlike jitter standard tests exist. For example a full scale 20kHz LPCM signal on a CD.