Integrated Amp Reviews

I measured the Roksan Kandy K2 BT using my Audio Precision SYS2722 system (see www.ap.com, and the January 2008 "As We See It"). Before performing any measurements on an amplifier, I run it for an hour with both channels operating at one-third its specified maximum power into 8 ohms. Like the earlier Kandy K2, reviewed by Art Dudley in May 2010, the K2 BT got very hot after 30 minutes, at which point I cut short the conditioning.

The first test I usually perform with an amplifier after this preconditioning period is to examine its behavior with squarewaves. The original K2's reproduction of a 10kHz squarewave was excellent, with a very short risetime (see fig.4 here). By contrast, the K2 BT's reproduction of a 10kHz squarewave (fig.1) had long risetimes, equating to a limited ultrasonic response. It turned out that the Roksan's frequency response depends to a larger extent than usual on its volume-control setting. Fig.2 shows the amplifier's response with the volume control set to 9 o'clock (left channel green, right gray), 12 o'clock (left blue, right red), and 3 o'clock (left cyan, right magenta). At the lowest setting of the control, the output is down by 3dB at 95kHz, but with it set to 12 o'clock, which is where I'd measured the squarewave response, the –3dB point has moved down to 24kHz. Increasing the volume past that point increased the ultrasonic bandwidth, and at 3 o'clock the –3dB point was 35kHz, while with the control set to its maximum (fig.3, blue and red traces), the –3dB output lay at 115kHz. Note, too, that the channel matching also changes with the setting of the volume control, the maximum difference between the channel gains (0.35dB in favor of the right channel) occurring with the volume control set to 12 o'clock.

But it's the Bluetooth input that distinguishes this version of the Kandy from the one that AD reviewed, and I examined the performance of the K2 BT's Bluetooth input using both an iPad 2, which uses the default AAC codec, and my 2012 MacBook Pro, which has the better-performing aptX codec (all MacBooks with OS 10.6.5 or later have aptX). Because the Bluetooth connection uses a lossy codec, measuring its performance is no straightforward task. (I offered a primer on assessing Bluetooth behavior in my review of the Arcam rBlink Bluetooth D/A processor in March 2014.)

Playing files with iTunes with its volume control set to the maximum, the Roksan's analog output waveform, assessed at the preamplifier output jacks, was severely clipped. I had to back off the MacBook's volume control to approximately 90% before the K2 BT's output waveform was free from clipping; at that point, the level was 2.3V. So if you use the Roksan's BT input, keep your source's volume control down. However, when I sent the K2 BT dithered 16-bit data representing a 1kHz tone at –90dBFS, there was no 1kHz content in the spectrum of the amplifier's output. This was presumably because I'd set the laptop's volume control to 90%, to avoid clipping, which had dropped the tone below the 16-bit quantization floor. Resetting my laptop's volume control to its maximum and repeating the analysis gave a spectrum with a 1kHz tone at –90dBFS (fig.8). However, the noise floor is hashy-looking.

The Bluetooth connection is limited to sample rates of 44.1 and 48kHz and a bit depth of 16. The BT input's impulse response was that of a conventional, time-symmetrical, linear-phase reconstruction filter (not shown), while the frequency response with 44.1kHz data featured a gentle rolloff in the top audio octave that reached –1dB at 20kHz (fig.9). With a full-scale 1kHz tone (fig.10), the aptX codec's limited bit budget results in the noise floor rising below 7kHz, while with the gapped, multitone test signal that I discussed in the Arcam review (fig.11), the codec preserves as much resolution in the midrange and low treble, where human hearing is sensitive, compared with the top two octaves, where hearing sensitivity decreases. Tested with the 16-bit J-Test signal, which comprises a high-level tone at one-quarter the sample rate and an LSB-level, low-frequency squarewave, the rise in the noise floor in the top octaves masks both any jitter artifacts and the odd-order harmonics of the squarewave (fig.12).

As I wrote in the Arcam review, the aptX codec throws away absolute resolution in favor of preserving a random noise floor because this will be less annoying with music. But the true test of a lossy codec is to listen to it, which is what HR did, of course.—John Atkinson