We believe the Kassandra dac is currently one of the best audio sources available today and can only be matched by top analogue sources in timbre and musicality, but keeping well ahead in aspects of dynamic presentation and resolution.
While in the process of setting up and building the company, at the most early stages, the need of a true SOTA source was mandatory. You cannot claim to built speakers that aim to be the absolute best, top amplifiers designs and so on, if you do not have a true reference source. And I mean reference in the true context.
The source must be absolutely perfect
We began developing the Kassandra as I felt no source on the market would stand up to the task. There would always be a certain character, lack of dynamics, lack of musicality and true analog-like timbre and flow. And these shortcomings would project onto our efforts to create the best speaker-amplifier combination possible.
In the audio industry today, the market is in abandon with digital processors with a new dac presented every week, in a race of finding the best converter. A race of numbers, with THD of -110db, oversampling @ radio frequencies and 140db dynamic range. Yet, many audiophiles change dacs as soon as the next one arrives, without ever finding the one sounding “just” right.
In the process of development of our reference system, we were also searching for the best available converter. Yet no converter delivered the sound we expected, as a reference converter. Every dac presented an artificial feel to every recording, and almost all presented compressed dynamics and poor flow of music, easily noticeable with uncompressed recordings and ultra high sensitivity horn systems. So we began designing the best possible digital processor.
Delta sigma dacs are todays standard for digital processors. Not because they provide better audio quality, rather because of availability and ease of implementation of digital filters. Filters that became more and more complex, only because of the artefacts created with delta sigma modulation.
The Kassandra’s prototype was developed very early, not as a commercially available design, but as a cost no object lab tool, my personal reference tool for speaker and amplifier development. The decision to be put it in production and to be commercially available was decided at a much later stage. We found since then, with many new converters launching daily with impressive specs on paper, they fall short of what we call analog sound. They always presented that electronic signature that gives digital reproduction a bad name.
As I like to say, my aim was making a true analog source, which would take
Ones and Zeros as input, and I feel we succeeded.
The ladder dac is a resistor network switched by a N number of switches, N being the bit depth. It is a passive sort of speak procedure, whereas DS technology, while still developing and moving forward, is a completely different kind of conversion procedure. DS converters were developed because it is many times cheaper and easier for the IC company to make a DS modulator, than to manufacture an R2R dac which includes costly procedures like laser trimming resistors to an accuracy of 1/128000 it’s value, the minimal accuracy needed to create a linear R2R converter.
Creating analog signal from noise shaping (as in DS conversion) sounds counter intuitive. While the very complex, very high order filters that are implemented to reconstruct the analog signal, from a burst of noise, makes a good job creating very good specs on paper. However the very high complexity of the filters and the resultant very high energy – several KHz-MHz noise present, is what gives all DS converters their distinctive sonic signature, a sonic attribute often given the term ‘digital sound’.
Back in R2R principles, the individual resistors must be laser trimmed to a incredible accuracy, while making extremely high accuracy measurements, a process unfortunately not available to small batch manufacturers. It is practically impossible to achieve the accuracy of integrated circuits with discrete resistors. Even the random solder resistance will ruin the low level linearity. We chose to use an R2R IC, in our case the classic Analogue Devices AD1865N-K, a much praised chip which we believe is the best sounding IC ever made. Many DACs use the specific IC converter in the usual “data sheet application” circuit, just adding their analog stage and, voila.
We choose to design a converter that uses a number of IC R2R converters as components to a completely new converter system. The AD1865N-K has a very straight forward data handling logic, and does not process the data stream in any way. This gave us the freedom to fully exploit the IC on our converter system. For example, the input data are latched and directly refreshes the resistor network, with absolutely no additional complex logic or data handling/processing. This way we have full control over the resistor network to use it as part of our converter system. Our Super Clock circuit directly re-clocks and drives the “refresh” signal, so absolutely no additional jitter is induced in the conversion. This is not possible by any other IC converter, as there is always internal logic and data handling, vastly deteriorating jitter performance, even if the clock input is close to perfect.
The massive banks of the IC converters act as parallel switched resistor ladder converters, thus cancelling the deviation of the actual resistor values vs the theoretical. If the deviation of the resistor value from ideal is Gaussian (which it is), as the number of parallel resistor networks increases, the deviation from ideal is driven to Zero. This is why paralleling R2R converters improves linearity. Noise figures are improved on the same principles.
Paralleling of converters improves the measurable parameters as well as the sonic characteristics of the converter. The total Signal to noise ratio is improved by doubling the number of converters, and linearity is also improved along with the dynamic range and channel separation. Sonic wise, paralleling converters elevate performance in another level in all aspects significant in high end audio. Each doubling of the converters in parallel sets a new level of performance. Dynamics and micro-details improve proportionally, so you could easily say that it is another converter altogether.
The Kassandra converter is working in complementary mode. Each channel is consists of two converter banks which are “mirrored”, perfectly synchronized and matched. This further improves low level linearity(the linearity near “zero” value), as well lowering noise.
We use many techniques regarding lowering and cancelling jitter, in both passive and active ways. Bouncing signals, overshoots, low rise times etc greatly increase jitter. We solve this by using special driving circuits and carefully tuned digital line terminations for all our digital signals.
Crosstalk between digital lines, ground noise and power supply noise can be a major source of jitter. Thus our DAC has extensive use of 35 LC filters implemented using RF chokes and high speed capacitors to decouple all digital circuits and ICs etc from the power rails. These are extremely effective in cancelling out any form of PSU sourced jitter. Thus using high loss RF chokes and high speed capacitors to all ICs, power rails and logic circuits have been used to solve this fairly complex problem. Jitter originated in USB and S/PDIF / Toslink sources are eliminated right at the conversion enable line, or the “refresh” signal of the converters.
Having an IC that has no complex data handling logic and no multi stage data logic, the precisely timed signal resolves in a extremely accurate jitter free conversion. In our opinion, this is not possible with any other type of converter system.
The Super Clock is also used upstream for re-clocking the XMOS asynchronous USB controller, as well as the SPDIF receiver. With the internal clock, there is no need for a word clock input. Jitter is very audible, and I felt that the dichognomy must end. That is why the digital input re-clocking circuits are by-passable on the fly, so every audiophile can make a AB comparison with the turn of a knob situated on the top of the Kassandra case.
• 16 R2R converters per channel, complimentary current output using the top grade Analogue Devices AD1865N-K with 8 converters per bank, 16 per channel.
• Eight discrete ultra-low-noise regulators for the 4 converter banks.
• Extensive local decoupling,using tuned LC filters.
• Over engineered power supplies, power input filters.
• Transformer I/V conversion with custom wideband transformers, balanced current to single ended voltage conversion.
• Internal Super-Clock by-passable on the fly. , triple regulated supply.
• Transformer loaded super tube output stage using the E280F tube. 5:1/10:1 step down transformer & double choke filtered supply.
Output range: – 30Vpp output @ 0db 10Vrms (5Vrms on -6db setting)
User selectable output: 10v @50ohms or 5v @12ohms
Output: 2 x RCA -True Balanced output standard. Floating RCA output switchable(ground loop resolver)
Output impedance: ~50ohms (balanced and SE output), 12ohm at -6db setting
USB input: up to 24/384KHz
Jitter attenuation: down to femtosec level
Dimensions: 540mm W x 520mm D x 165mm H