As can be seen from the pictures, Ayre retained the same chassis as in the original version, with the sole exception of the addition of the term "DSD" in the logo. This means that you are facing a very simple machine without any buttons (and with the power switch on the back panel, a practice that, to my opinion, is not entirely correct but regularly adopted) and a large, easy to read display which gives information about the sample rate the input has locked on.
The back panel is also simple, including two pairs of analog outputs (single ended and balanced), the only input, which is the USB port and a block of DIP switches that control some of the functions of the device such as the auto-On, the brightness of the screen (the user can simply put it off via the switch, the continuous adjustment requires the existence of another Ayre device in the system, connected through AyreLink bus), the type of connection (Class 1/2) and the digital filter to be used.

Inside its chassis, the device conceals an ordered and top quality construction that reveals much more detail about the changes that have been made than its external looks.
Management of the signal through the USB port is assigned to an XMOS chip solution, as in the previous edition, providing asynchronous data transfer and support for the DoP (DSP over PCM) standard that allows the converter to stream DSD 64x (2.8MHz) files. The company has already announced a small (software level) upgrade that allows DSD 128x playback but this was not available at the time of the review. PCM streaming maximum sample rate remains at the usual 24bit/192kHz and for sample rates above 96kHz the installation of a driver is required, as one would expect, for Windows machines (ASIO for the case of DSD). As in the previous version, the QB-9^DSD uses Gordon Rankin's Streamlength technology from the asynchronous transfer.

The stream from the USB interface is processed by a Xilinx FPGA obviously the center of the unit's entire digital signal processing. The fact that there is no conventional S/P digital interface chip here, partially justifies Ayre's design approach to not include an S/PDIF input, something that the majority of QB-9's users will probably miss. Digital filter implementation in the QB-9^DSD does not seem to have undergone changes (at least Ayre says nothing relevant). It is, therefore, the same minimum-phase design Ayre introduced with the original QB-9. It is founded on the work of Algol 's Peter Craven ("Antialias Filters and System Transient Response at High Sample Rates") which proposes the use of specially designed low pass filters (known as apodizing filters) to improve the performance of a digital system in the time domain. Ayre explains in detail the course from Craven's suggestions to the practical filter used in QB-9 here, the main idea being the optimal choice between the inevitable pre-ringing of a linear phase filter with very steep slope in the transition region and the standard ringing of a minimum phase filter with a milder slope, which appears after the main impulse response, a much more “natural” behavior for any system. According to Ayre, Craven and Meridian's Bob Stuart, pre-ringing (i.e the existence of transient phenomena before the filter main impulse response that occurs only in digital systems as a byproduct of the processing of previous samples) generates unnatural audible effects in contrast with conventional ringing (i.e the existence of a delay in damping of the impulse response) which is more "natural" and more benign. Stuart gives a better explanation on this issue arguing that if the ringing is small, then it is not perceived by the listener since psychoacoustic masking phenomena are strong enough to cover it. At the level of implementation of all these, QB-9^DSD offers two digital filter options: One with a higher cutoff frequency and a high slope called the "Measure" mode and one with a lower slope and a lower cutoff frequency called the "Listen" mode (a pun intended, probably). Apart from the different behavior in the frequency domain, the two filters differ in the time domain too, as one expects, the "Measure" being slower in its damping and with a more prevalent ringing part.

Clocking of the device has also changed. As in the previous version, two different time base circuits are used to obtain the basic sample rate (44.1kHz/48kHz) but the company states they now offer lower phase noise (i.e a lower jitter figure) and are operated at twice the frequency compared to those of the previous model, an approach that ensures better timing conditions for the converter chip that follows. In place of this key component (replacing Burr Brown DSD1796 of the previous version), Ayre has chosen the ES9016S, one of the top converters from ESS. The 9016 is an 8-channel design therefore, although Ayre does not mention anything about that, one can assume that some form of parallel topology (four 9016 channels per QB-9 channel) is used here.
The analog stage that follows is based on the circuit used in the original QB-9. This, according to details given by Ayre, is a balanced design with discrete semiconductors and top quality components without feedback. The company states that some "changes have been made" without becoming more clear on this point. Changes have been made also to the power supply which now includes a separate line for the USB interface (the previous version used the bus power supply), a choice that ensures very low noise levels, Ayre says. Additionally, the input stage of the USB port is optically isolated for even lower noise (especially the part that originates from the computer).

Previous | Next