Figure 3

Oversampling makes use of interpolation to create the extra samples. However, interpolation alone between two consecutive samples is not sufficient to attenuate the sampled images at FS through to 7 x FS, hence the reason for digital filtering. The digital filter makes use of many more samples to perform the interpolation, the more samples it can use, the better the interpolation and attenuation of the sampled images. The attenuation of these images is very important for high performance audio reproduction. This is why the DAC20 uses a PMD100 eight times oversampling filter which offers outstanding performance in terms of its passband ripple (< 0.0001 dB) and stopband attenuation (> 120 dB). The filter also has the ability to decode HDCD encoded CDs, thus allowing the full benefit of increased dynamic range offered by these CDs to be enjoyed. HDCD HDCD is an encoding and decoding process developed by Pacific Microsonics to provide increased amplitude resolution and to correct for sampling rate limitations. It is compatible with the CD "Red Book" standard, thus allowing encoded CDs to be played on equipment not containing HDCD decoding, but still offering some of the benefits of improved fidelity. However, for the full benefits of HDCD, the use of a decoder is essential. This is another reason why the DAC20 uses the PMD100 digital filter which contains HDCD decoding. The DAC20 will therefore allow the full performance of the growing number of HDCD encoded CDs to be enjoyed. HDCD encoded CDs contain a hidden code that is detected by the PMD100 for automatically switching in the decoder. Since this process is automatic, there are no controls for HDCD on the front of the DAC20, but there is an indicator to show when HDCD material is being decoded. The peak output level from the DAC20 is 6 dB higher for HDCD encoded CDs. This is due to the increased dynamic range offered by HDCD. Digital to Analogue Conversion The Digital to Analogue converter stage is constructed around the SAA7350 and TDA1547 from Philips. The SAA7350 is used to up sample and noise shape the digital audio data stream, producing a 1-bit Pulse Density Modulated (PDM) output signal for conversion back to analogue by the 1-bit DAC. It takes in data from the digital filter at 8 x FS, and then outputs it at 192 x FS. This oversampled data stream is then carefully routed to the TDA1547 1-bit DAC. The clock for the DAC is taken directly from the clean master clock, which is again carefully routed away from digital audio data tracks. The layout of the Digital to Analogue Converter stage is as important as the design itself, due to the high frequency content of both data and clock. Special RF layout techniques are used to maintain converter accuracy and achieve optimum performance. The attention to detail also ensures that the full benefits of single bit DAC are obtained. The voltage reference to the 1-bit DAC is derived from a forward biased LED. This produces significantly less noise than a standard voltage reference diode which relies upon reverse bias breakdown. The reference voltage from the LED is buffered and heavily filtered to provide a low impedance drive across a wide frequency range, up into the MHz region. This again is essential for accurate conversion. Noise Shaping and 1-Bit DAC The process of oversampling and how it can be used to reduce the word length has been described briefly. For each doubling of the sampling frequency, one bit can be dropped from the word length. If this process is taken far enough the word length can be reduced to a single bit and a 1-bit DAC can be used to convert the samples back into analogue. Unfortunately, to reduce a 16-bit word down using this process would require an oversampling rate up in the GHz region. Since this is totally impractical, a compromise is made and a process called noise-shaping is used to get down to a single bit at sensible oversampling rates. Present noise shapers run at rates from 64 x up to 192 x FS. The higher the sampling rate the better, for obvious reasons. This is why the DAC20 uses an SAA7350 which runs at 192 x FS. Noise-shaping is used to reduce the noise in the baseband, which has resulted from the inability to oversample at a high enough rate, by moving it up to frequencies outside of the baseband. The SAA7350 uses 3rd order noise-shaping to achieve this objective. The noise shaped 1-bit waveform is processed in the DAC20 by the TDA1547 1-bit DAC. This part has been chosen for its excellent performance and linearity. Prerequisites for Accurate Digital to Analogue Conversion In order for the original signal to be reproduced accurately from the digital data, it is essential that the data is error free and the clock used for the conversion process is precisely timed. Clock purity is a particular area where great attention to detail is paid in all TAG McLaren Audio digital audio products. It ensures that the reproduced audio is precise in every detail, making the experience life-like. It is well known that the digital audio interface has its limitations, and that it is the main jitter contributor in outboard DACs. The reason for this is that the recovered clock, extracted from the data stream using a phase locked loop (PLL), tends to be modulated by the digital audio data samples. One would argue that if this is the case, then single box CD players must be better. This would certainly be true if nothing was done to remove the jitter from the clock. Jitter and Jitter Reduction The term timing jitter is loosely used to describe timing errors associated with digital signal transitions relative to their ideal positions. For clocks, timing jitter can be considered as a type of phase or frequency modulation. The clock recovered from the incoming data stream tends to be modulated by the data, an inherent weakness of the AES/EBU and SPDIF interface. This correlated modulation also tends to be worse for low level audio signals, where all the data bits tend to be changing from all ones to all zeros and vice versa. The DAC20 contains three PLLs for clock recovery, although generally, we only talk about two. The first, which is part of the digital audio receiver (CS8412), has a high gain VCO and a fairly wide closed loop bandwidth. The wide bandwidth is essential for correct operation of the receiver, but does not provide for good jitter rejection. The second PLL has a lower closed loop bandwidth and performs a given amount of jitter rejection, whilst at the same time up-converting the 256 x FS clock to 384 x FS. The clock from this PLL is used for data conversion when the Master Clock PLL cannot lock. The third PLL, which generates the Master Clock, is where the ultimate clock performance is achieved using Voltage Controlled Crystal Oscillators. Master Clock PLL The DAC20 is designed for optimum performance at each of the three standard sampling rates; 32, 44.1, and 48 kHz. At these sampling rates the DAC20 will operate in crystal lock mode using the Master Clock PLL. The block diagram, in figure 4, shows the basic construction of the master clock phase locked loop. Whilst the diagram only shows one crystal oscillator, there are in fact three, one for each of the sampling frequencies. The oscillators run at 384 x the nominal sampling rates. Only one oscillator is enabled at any one time to avoid problems of inter-modulation. The loop comprises of a comparator, loop filter, voltage controlled crystal oscillator, and divider. Although not shown in the diagram, the divider has in fact tap positions for other multiples of the basic FS sampling rate.

Figure 4

The crystal oscillators are designed to have pulling ranges of around ± 150 ppm, and can be locked to digital sources whose oscillators are in error by up to ± 100 ppm. It should be noted that all high quality digital audio sources will be well within these limits. A key feature of the oscillators is their low noise, which is essential for minimum jitter. The loop filter has been designed for a 3 dB closed loop bandwidth of around 12 Hz to ensure that jitter above this frequency is attenuated. The loop bandwidth also determines the lock time, and is the reason why the DAC20 can take up to half a second for its oscillators to lock. The performance of the loop can easily be seen by comparing the CS4812 256 x FS clock to the 384 x FS master clock using an RF spectrum analyser (see plots in figures 5 and 6). The audio test signal used for the comparison was a - 90 dB full scale 500 Hz sinewave with no dither. This was chosen because it is the lower level signals that produce most jitter on the digital interface and most test CDs contain this signal with and without dither. The reason for lower level signals being worse is that the majority of the bits in each audio sample are changing between all ones and all zeros.

Figure 5 - CS4812 256 x FS Clock

Figure 6 - 384 x FS Master Clock

Analogue Reconstruction Filters The DAC20 uses reconstruction filters to remove the high frequency digital images. Whilst these images are outside of the human hearing frequency range, and thus cannot be heard, it is still essential that they are removed. This is because other equipment in the audio path (preamplifier and power amplifier) could, by means of inter-modulation, introduce spurious signals into the baseband, thus ruining the reproduction. The filters are a five pole design offering good attenuation in the stopband whilst maintaining a constant group delay (linear phase) in the passband. The first stage makes use of OPA627s from Burr Brown, whilst the remaining stages make use of OPA134s, also from Burr Brown. High accuracy, low temperature coefficient, Vishay metal film resistors are also used, along with Wima FKP2 (polypropylene) capacitors. These components ensure both accurate filtering characteristics and audiophile performance. The output stage is direct coupled, using a servo to remove DC offsets. The servo provides a low frequency break point of around 0.1Hz, which is essential for maintaining the phase information of transient signals. The servo amplifier is an OPA627, the same as is used in the first stage of the filter. The final output stage provides a low impedance (100R) drive to the phono output sockets, thus allowing long cable lengths to be driven. The outputs are also stable for all capacitive loads, making them suitable for use with all types of audiophile interconnects.