Grafik Deutsch
Grafik English

Grafik Download as PDF





DeepColor - Ultra Low Jitter Clock for 24 Bit Audio




Including a brief Introduction to Design Guidelines for High Resolution Audio Clock Applications

by Ing. Michael Gerstgrasser










INTRO




Sometimes on a beautiful day at certain seasons we get touched by the purity and depth of the colors of the landscape we live in, suddenly experiencing what we have known all our lives as something exceptionally new and outstanding.

Sometimes we enjoy the same emotional excitement with fine audio gear, letting us dive into the flow of music we love and feel deeply the suspense and relaxation of harmony and rhythm, establishing a sort of intimate communication with the performing musicians overlaid by the ideas of composers and recording engineers.

As often with audio devices, there is a strong desire to find out the ingredients and technical specs that are relevant for what excites and touches us most in reproducing music like "LIVE".

Many electronic design aspects have been said to be the "really important" ones in the past, but not one has been proven to be "the one" to rely on only until now. Not even the most extended set of measurable specs seems to be able to define what someone judges to be "right".
All thats left for approaching any aspect related to the improvement of audio gear is to search for an electronic design optimum and have a look at what may have been overlooked or at least not been realised to be sonic relevant.

When the 16 Bit / 44.1 kHz standard for CDs, as the very first widely spread digital audio format for common use, was specified in the early 80ies of the last millennium, an expected 20Hz to 20kHz human hearing ability together with the sampling theorem stated by Harry Nyquist, Claude Elwood Shannon, Vladimir Kotelnikov and John Macnaughten Whittaker roughly eight decades ago, were taken as the basis to calculate the theoretical requirements.
Whereas the resolution in terms of quantisation always has been in the focus of general discussions reflected by several different approaches in AD/DA conversion the more subtle issues of resolution in the time domain were discussed much less.

Within the given limitations of the 16 Bit / 44.1 kHz CD format, some considerable improvements developed over time. I recall my listening experience of my brothers very first and brute sounding Philips CD-player against the sweetness and accuracy of his Sondek turntable at that time. The standard quality we get now in digital consumer audio both equipment-wise and due to refined mastering has largely improved since then.

Though, surprisingly much room for audible improvement still can be found, when concentrating on the behaviour of the crystal oscillators implemented in any kind of digital audio device.
Among several configurations the oscillator arrangement found by Edwin H. Colpitts in the early 1900's has been proven to perform very well in such applications.
Several variants of the Colpitts circuit exist, using tubes, BJTs, CMOS or FETs and others as an active device. Despite choosing any kind of favourite circuit, many considerations have to be taken into account and balanced out carefully to operate a chosen arrangement at its sweetest spot and to obtain the very best overall performance at the end of pipe.

Extended measuring effort, theoretical investigation and circuit simulation was done during development of the "DeepColor" clock design.
Though I find SPICE simulation here much more meaningful than with analogue audio circuits and optimising sound performance develops rather straight forward aided by this great tool for electronic circuit analysis, a lot of prototype listening sessions within an appropriate sensitive 24 Bit audio chain were necessary in order not to get trapped or mislead by excellent figures only.







How does it Sound ?



Though you should not expect any changes in tonality as frequency response is not affected, simply everything will sound different, from a tender swinging triangle down to the mightiness of the deepest notes your chain is capable of.

You will regain any transient resolution lost, all character and color provided by your records. Any uncertainties are wiped away, a stable and reliable image is gained that makes it easier to follow any detail while keeping track to the whole, regardless of how complex or minimalistic the music you play is.
In the higher region a certain kind of sharpness and harshness re-integrates into sound as very fast transients, enhancing "3D" focusing. There is sweetness from cymbals, astonishing attack with percussion and intimate clarity from voices. In the low region you regain the highest definition possible and a kind of transparency, you would not have expected there, transcending into a virtually unlimited sense of room perception. Improvement of  "3D" perception is even more pronounced at out of centre listening positions. 
Any sound and acoustical event is highly distinguishable by its purer and richer colors and a timing of the tune that does not lay back nor get into any hurry providing a strong feeling to stay "just in time", whatever rhythm or mood your music follows.

The above attributes are usually expected in describing speaker behaviour and are most likely associated with quite some investment. Regardless of how perfect or sweet speakers may sound, it's normally quite a challenge to balance analytical and emotional aspects there, whereas this is in no way conflictive for jitter improvement. Treating jitter behaviour at the front end of a chain, you always gain - or lose - in detail as well as overall. A very well designed clock lets you enjoy your records even more authentic, precise and with all the beauty music can bring to us.
Its pretty much the same effect when using a really nice lens instead of a blurry one when taking pictures.



Basically absolute EVERYTHING in music and sound is related to time. What was of no concern with analogue audio besides wow and flutter of turntables and band machines, becomes a completely new significance in digital audio in terms of sound performance all the way from close to DC over the audio band up to RF frequencies. To pioneer the challenge of full 24 bit audio clock resolution, the DeepColor clock was designed for and - believe me or not - it was as much joy for me to learn that much about unexpected sonic impacts of electrical and physical phenomena at this specific application as it is for me to enjoy the result now.

At the very end it simply is plain fun and joy to listen to whatever you like, turning on your heat or let you slip into magic meditation.









AIM, TECHNICAL DESCRIPTION, CIRCUIT DISCUSSION and DESIGN GUIDE LINES



Low audio band and sub audio band jitter are not recognised to be of special sonic relevance yet. My findings are different from mainstream opinion here and with the DeepColor clock you will be able to prove this yourself.  Two examples to illustrate jitter that can be found on consumer and pro-audio device:



Grafik  Grafik 
Fig. 1 Random low frequency noise on ONYX 400F
sound card  with internal clock used.
Fig. 2 Jitter of  PHILIPS CDR 795 PLAYER in ULTRA-ZOOM 
measured with 32k and 256k FFT points (0,2 Hz resolution, 
red graph)


What can be seen here is that considerable amounts of audible jitter can be found in consumer and pro-audio gear. 


To put the superior performance of the "DeepColor" clock design into perspective, comparison with two other after market super clock designs acknowledged to work fine, is provided as far as useful. This schematics are free for non commercial use and can be found to download under :
LC Audio XO3 schematic
Elso Kwak clock ver7 schematic



Rather than only claiming superiority - as seen from most other clock manufacturers - this compilation focuses on basic technical relations for a better overall understanding on the very interesting topic of jitter in audio applications. Though there are many figures presented below, it's suggested, to bear in mind the overall picture as specific figures may vary from sample to sample and SPICE modeling also never is perfect.

Two major statements always have to be kept in mind when dealing with jitter in audio audio applications.

First of all, a very good clock signal especially tailored for the AD / DA converter is always preferable to an external clock, avoiding the additional sources of PLL jitter, cable jitter and others.
Secondly, jitter does not affect sound performance during digital processing. Jitter only comes into play when converting signals from the analog to the digital domain and vice versa. The only exception here are sample rate converters that mimic this task back and forth with the help of digital algorithms. For SRCs the input clock jitter AND the output clock jitter contribute to the overall degrading of the digitally coded audio signal leaving the SRC.

Having said this, its easy to understand that following applications benefit most from an ultra low jitter module like the DeepColor clock :
- AD converters in any kind of device
- DA converters in any kind of device
- SRCs in any kind of device
- Clock distribution systems (in order to sync a whole digital chain with the device of choice rather then just outputting a so called reference clock)







CRYSTAL SIGNAL PURITY and OSCILLATOR SUPPLY CURRENT



The first thing important to consider of a quartz oscillator circuit may be the sinus signal purity of the crystal in the oscillator itself.
Due to finite impedance and certain hardly predictable behaviour of power supply circuits at higher frequencies, the current drawn by the active device also has unexpected audible relevant impact.
It modulates the effective supply voltage at side band frequencies above the centre frequency and sometimes even down all the audio band. Because the correlation with the centre frequency never is 100%, this AC voltage drop may rather be seen as a kind of inducted supply noise.
Depending on circuit layout, AC magnitude and frequency spectrum involved, the current drawn by the active device may therefore easily overlay and effectively degrade noise performance of the supply by several orders of magnitude.



Grafik  Grafik  Grafik 
Fig. 3 DeepColor Clock Signal Purity and AC Power Supply Current Fig. 4 BJT Clock Clock Signal Purity and AC Power Supply Current Fig. 5 FET Clock Clock Signal Purity and AC Power Supply Current


What can be seen here is that the DeepColor clock operates the crystal at a very clean sine wave oscillation and that the power supply interaction is greatly reduced by a factor of more than ten remaining a smooth signal here as well.






OSCILLATOR POWER SUPPLY REJECTION





Taking the output amplitude of the oscillator as a reference, it is useful to calculate the effect of variations in supply voltage with respect to the output signal.
What in data sheets commonly is referred to as the PSRR figure, is normalised to the sine wave output of the oscillator here and reflects directly how much the sine wave output signal of a certain oscillator circuit is degraded by any kind of noise or line and load regulation of its supply voltage. Any remaining distortion signal at this point translates directly to clock jitter via some complicated maths.
Though providing ultra low jitter figures got common use for advertisement reasons, it must be stressed that clock jitter figures are absolutely useless if not specified for a certain frequency or bandwidth.

In case of sine wave distortion signals below the Nyquist frequency, one will get side band lobes at a distance related to the frequency of the distortion signal and depending on the level. With random noise as distortion one will get a raise in noise level depending on band width and level.

Advanced high order filtering of the oscillator sine wave output signal of the DeepColor clock provides exceptional noise distortion rejection at the most critical point of the clock circuit. One has to keep in mind that phase jitter will occur with all signals that overlay the oscillator sinus and will translate audible relevant even down to the sub audio band.




Grafik  Grafik  Grafik 
Fig. 6 DeepColor Clock Oscillator PSRR Fig. 7 BJT Clock Oscillator PSRR Fig. 8 FET Clock Oscillator PSRR


What can be seen here is that the DeepColor clock provides superb audio band and sub audio band power supply rejection, reducing any jitter prone signal distortion by oscillator supply imperfections at the point of the comparator threshold to negligible amounts. All supply signals are phase locked to show a worst case scenario.






SIGNAL and NOISE SHAPING




Converting the sine wave from the oscillator to a square wave clock signal can easily be done by a high speed comparator. There are brick wall limitations to jitter performance though at this point that can not be overcome by any design tricks. Few but very basic laws determine random jitter performance, which lead naturally to general design rules.

Assuming the rest of the circuit to be ideal, the minimal random phase jitter achievable depends on the peak to peak noise of the signal and the comparators input voltage noise in relation to the signal's slew rate at the point of threshold. This jitter is basically wide band and sometimes referred to as random jitter RJ in ultra high speed comparator data sheets. RJ figures have to be interpreted correctly, as they depend upon the slew rate of the signal. Very low RJ figures achievable with GHz signals are of no meaning in an audio clock application with limitations to some tenth MHz.

The mechanism of random phase jitter, actually representing the uncertainties about the exact time for toggling from low to high and back again, best can be understood by a signal traveling a peak to peak noise floor at a given speed. Simplyfied random jitter can be seen as the equivalent to noise in the analog domain. One difference between random jitter and analog noise is that uncorrelated noise at the point of threshold does sum up as jitter rather linear than in quadratic terms as it would be with analog noise.




Grafik  Grafik  Grafik 
Fig. 9 noisy signal passing a noisy threshold Fig. 10 ideal signal passing a noisy threshold Fig. 11 noisy signal passing an ideal threshold


What can be seen here is that random phase jitter is added by the noise floor of the oscillator signal and by the noise floor of the comparator input in relation to the speed of passing through these noise floors



Adding hysteresis to the comparator may help in preventing ultra high frequency oscillation under certain conditions, but can not further improve jitter performance. What can be done to optimise audible performance is, to exclude jitter frequencies in the audio band and sub audio band as far as possible. This is accomplished for the DeepColor clock by inserting a high order filter between oscillator and comparator. A low impedance, low noise, high PSRR output stage is implemented in the DeepColor clock's gain block to drive this filter at an unusual high signal level.

By the way, Fig 11 allows an intuitive understanding of phase noise figures. At the point of threshold noise in the y - axis has basically the same effect and cannot be distinguished from noise in the x - axis. What was considered as noise in the y -axis until now can therefor be seen and expressed also as noise in the x- axis which means variations in frequency. Normalised as a fraction of the oscillator frequency phase noise figures are convenient when comparing the effects of circuit design at different oscillator frequencies




Grafik  Grafik  Grafik 
Fig. 12 DeepColor oscillator output noise Fig. 13 BJT Clock oscillator output noise Fig. 14 FET Clock oscillator output noise


Y-Axis scale is V/Hz½.
What can be seen here is that the DeepColor clock oscillator has negligible low output noise up to 1 MHz.





Grafik  Grafik  Grafik 
Fig. 15 DeepColor oscillator RJp-p Fig. 16 BJT Clock oscillator RJp-p Fig. 17 FET Clock oscillator RJp-p


What can be seen here is:
Calculating a 10 Vpp signal with an oscillator output noise around 500 nV/Hz½ from the DeepColor clock will translate to a theoretically minimal achievable random jitter of around 10 ps at 22 MHz. This amount of random jitter is limited to the upper MHz region falling rapidly to ultra low values for any frequencies below 2 MHz. Targeting a tough 500 fs, the DeepColor clock gives a considerable jitter improvement over existing designs by a factor of roughly 20 to 200 throughout the audio and sub audio band!
The DeepColor clock design would benefit from a noiseless (!) comparator dropping RJp-p below 1 MHz wide band into cellar.

Calculating a 80 mVpp signal with an oscillator output noise around 2,2 nV/Hz½ from the BJT clock example above will translate to a theoretically minimal achievable random jitter of about 90 ps. RJ is distributed equally over frequency. This design would benefit from a noiseless (!) comparator dropping RJp-p to around 25 ps wide band.

Calculating a 700 mVpp signal with an oscillator output noise around 1 nV/Hz½ from the FET clock example above will translate to a theoretically minimal achievable random jitter of about 10 ps at 22 MHz. At around 500 Hz the output noise of the oscillator is peaking to about 300 nV/Hz½ which translates to jitter around this frequency of about 380 ps falling to around 15 ps at 100 kHz. This design would benefit from a noiseless (!) comparator dropping RJp-p to around 2 ps at 22 MHz / 370 ps falling to 7 ps from 500 Hz to 100 kHz.





Above calculations assume a low comparator input voltage noise of 6 nV/Hz½, held constant here to ease maths. Measurements show that this is a fair value for some widely available real world high speed comparators. Obviously only intelligent use of standard filter techniques for the oscillator signal - like practiced in the DeepColor clock design - allow to benefit substantially from ultra low noise comparators having even lower input voltage noise.

It has to be kept in mind that all RJ figures calculated here represent absolute design limits only, one may or may not come close to, depending on a clever balance of ALL requirements necessary to reach the goal. Though the DeepColor clock is designed to perform in an outstanding way also in this aspect, the audible irrelevance of a one number specification was clearly outlined above which can not be stressed enough. The RJp-p figures are based on a roughly 22 MHz sine wave signal and 100 MHz band width, crest factor is assumed to be 6 for simplicity. Please note that as an optimistically assumption the noise floor of the threshold is considered to be constant over frequency throughout all calculations.



There is an other aspect worth to be look at, concerning the arrangement of filter components for achieving best results. A comparator input voltage noise is specified with the input shorted, requiring to keep the source impedance as low as possible - if above RJ figures should hold true.



Grafik  Grafik  Grafik 
Fig. 18 DeepColor Clock Output Impedance Fig. 19 BJT Clock Output Impedance Fig. 20 FET Clock Output Impedance


What can be seen here is that only the DeepColor clock provides sufficient low output impedance in the audio band and sub audio band, shunting the input impedance of the comparator very efficiently by a optimised filter topology. Therefore no further degrading of the comparator's input voltage noise in this frequency spectrum occurs. In other words, the DeepColor clock oscillator design is not compromising the noise performance of the comparator at all.


Regardless of how low the input voltage noise comparators may become available in future, the DeepClock's superior design concept proves to be outstanding also in this aspect.

There are other jitter components determined by variations in power supply, signal purity, interfacing and others that may add to the total amount of phase jitter in a clock and these ones also have to be kept under tight control by circuit design, layout and proper interfacing.






Top Notch / Active Interfacing / Time Reference




The low impedance, high voltage output of the DeepColor clock's oscillator circuit allows for driving the load directly or via cable. This is a big advantage as only a single frequency sine wave has to be transmitted which saves a lot of cable, EMI and termination troubles.
The very small receiver unit can be mounted on top of the AD / DA converter and the DeepColor clock itself can be housed elsewhere. Two choices can be made here. Either a receiver unit that makes use of a high speed comparator with a high order filter in front of it or a receiver unit containing only the the high order filter and a sine wave clipping circuit.

The receiver module with the sine wave clipping circuit has the advantage of feeding the AD / DA converter threshold directly and without any additional noise source in between. The receiver module with the comparator circuit has the advantage of a slightly higher slew rate output.
What will result in lower jitter therefor depends on the noise level of the clock input of the AD / DA converter. The trade offs between speed and noise are outlined above in detail, though chip manufacturers normally don't specify the performance at the clock input of their AD / DA device.





Grafik 
Fig. 21 DeepColor Clock top notch RJp-p


What can be seen here is that the DeepColor clock design allows for virtually loss free interfacing.
Calculating 1 nV/Hz1/2 for the threshold of the AD / DA converter and using the clipped sine wave circuit receiver unit will give a RJp-p directly at the clock input of the AD / DA converter of slightly below 90 fs within the audio and sub audio band. Not that bad indeed!



What also can be learned here is that no matter if you use as a time reference a standard quartz crystal or atomic clocks - which in fact are capable to provide several orders of magnitude better long time stability << 1 Hz - the random jitter within the audio band and sub audio band is heavily degraded even by the very simple task of transforming a sine wave into a square wave done by any pretty good comparator. Every complex signal processing - like PLLs for frequency shifting for example - will degrade random jitter figures even more.

This puts an interesting light on the fact that we seldom see RJ versus frequency plots, even for Euro 10 000.- rubidium audio clocks.
It also makes clear that only a widely unprocessed clock signal that is fed directly and with the appropriate frequency to the AD / DA / SRC device ensures best performance.
Furthermore it makes clear that when RJ considerations at the side of the clock unit are consequently pushed to the limit, its up to the AD / DA manufacturers to specify their units in more detail as they are becoming the dominant part with respect to RJ figures.

Room for a first step overall improvement is defined easily:
1.) push down the noise figure of the threshold
2.) push up the clock frequency
3.) push up the amplitude of the clock output
4.) filter out unwanted noise if the clock signal

What makes the DeepColor clock design such nice is that it will not limit further improvements there.





DeepColor CLOCK MODULE SUPPLY and BRAKE-IN TIME




Thanks to nested on board supply voltage stabilisation circuits not covered in detail here and the low power consumption of the DeepColor clock, almost any internal supply of the device to upgrade can be used. The module accepts voltages in the range of plus / minus 12V - 35V and draws a current of roughly 50 mA. If unregulated supply is used, ripple should not exceed about 50 mVrms as a rule of thumb.

A quiet external supply in general allows for better results in terms of timing, flow and room perception as described above.

The DeepColor clock does not make use of any exotic parts but rather relies on solid construction. But as with any sound device, you may allow some time to settle before judging. Half a week of continuous operation will be fair, any further changes will be very sublime and it's more likely that it will take several weeks to realise the full potential of the DeepColor clock rather than the clock's performance will change substantially.







Clock System Requirements for Digital Audio Chains




It is obvious that a word clock system is the only way for complex digital audio chains to maintain synchronous operation of several digital audio devices without accumulating clock jitter. Every device of a digital chain must therefor be driven from a single word clock frequency or from a clock frequency that is phase locked onto it. Otherwise samples will get lost and the audio signal will be corrupted. For convenience word clock frequencies are normally distributed as the standard sample rate frequency or multiples from that.
This implies that such word clock frequencies normally must be shifted by a frequency multiplier or divider prior to feeding the AD / DA chip. Complex clock signal processing will have heavily adverse effects on jitter performance as clearly outlined above.

Simply the only ( ! ) way out of this dilemma is to do it just the other way around. Use an ultra low jitter clock that generates exactly the right frequency for your device of interest and lock the word clock and the rest of the digital chain onto that. Remember that clock jitter has its importance during AD / DA conversion. As long as you stay in the digital domain, clock jitter does not affect sound quality.
The DeepColor clock provides a second output to allow for external phase locked frequency conversion to a 10 MHz reference or standard audio clock distributing frequencies.

Almost any professional clock distributing device allows to lock onto an external 10 MHz reference. Depending on whatever recording or playback is the critical task in your application, more intelligent designed word clock distributing devices provide even more flexible switching options to phase lock onto severals frequency reference inputs.

Things are simple as long as the border between analog and digital domain is crossed only once. An external DAC driven from a transport or from a DAW during playback or an external ADC driving a DAW during recording would be examples for that. In digital chains like that, there is no quality degrading limit for adding digital devices in between, as long as they all work at the same sample rate derived from one word clock.

In case of playback the word clock has to be locked onto the DA clock - in case of recording the word clock has to be locked onto the AD clock.
As a footnote it has to be said that for digital recording chains jitter accumulation isn't really a problem and word clock is of no benefit as long as there are no multiple branches for effect devices involved.

Things get more complicated with digital audio gear that make use of sample rate conversion.
In that context SRC devices are best understood and treated similar to any devices that perform a digital to analog plus an analog to digital conversion within the same shell. SRC's basically always have two clock inputs: one for the incoming signal - as if it were a DA converter - and an other one for the outgoing signal - as if it were an AD converter. You are happy if the SRC is inside the device at either end of your digital chain. An external DAC accepting sources with a wide variety of sample rates or just from a specific transport with a different sample rate would be an example for that. Such designs are sometimes named as upsampling DAC's.
Though inside the same shell with such upsampling DACs the jitter behaviour of the SRC and that of the DAC should be considered separately. The outgoing SRC clock and the DAC must be driven from the same ultra low jitter clock. This clock frequency is important only for that device and hence isn't the one the other devices in the digital audio chain must be locked onto.
A SRC is jitter sensitive to its incoming clock frequency as well as to its outgoing clock frequency by measuring the difference of those frequencies permanently . The incoming clock frequency of a SRC is commonly extracted from the incoming digital audio signal data stream. This means that the source device for the SRC has to be clocked with an ultra low jitter clock as well.

In this case the SRC's sourcing device clock frequency is the one the word clock has to be locked onto and that word clock must be distributed to any device that may be ahead in the digital chain.

In rare cases where jitter impacts are considered throughout a complete chain, the SRC may be fed with a data stream and a separate data clock.

In this case the SRC device has to be operated with an additional ultra low jitter clock at its data clock input. This clock frequency has to be fed to the sourcing device and is also the one the word clock has to be locked onto and that word clock must be distributed to any device that may be ahead in the digital chain.

For SRC devices in the middle of a digital chain multiple word clock frequencies have to be distributed accordingly to the above - even if they are basically of the same value.


One conclusion of the above is that SRC's aren't that well suited for jitter rejection - and there are other reasons for this as well.







Preserving Signal Integrity at 16 Bit and at 24 Bit




What is audible or not always has been and will be under question. But what can be calculated is the amount of jitter that does not corrupt AD / DA conversion. To put things into perspective, lets have a look on the time step accuracy needed, if we sample with full 16 Bit or 24 Bit resolution within the audio band. Lets generously assume the audio band to be 10 Hz to 100 kHz.

Imagine for a moment that we wouldn't sample at fixed time slices but every time the analoge audio signal moves one bit in amplitude. Let's further assume that the full scale input range of our imaginary AD converter is 2.5 Vpp. Now let's calculate for only one half bit in order to preserve signal integrity at various analog signal frequencies at the point of their maximum slew rate.


Analog Signal Frequency 24 Bit Time step (1/2 LSB at 2.5 Vpp FS) 16 Bit Time step (1/2 LSB at 2.5 Vpp FS)
100 kHz 100 fs 25 ps
1 kHz 10 ps 2.5 us
10 Hz 1 us 250 us

Fig. 22 Signal Integrity Demands at 16 Bit and at 24 Bit


What can be seen here is that 24 Bit audio is highly demanding with respect to the correct timing of the samples. Didn't these ingenious fellows tell us that a hundred years ago?

What seems to be well within reach for the DeepColor clock even for 24 bit / 100 kHz when operating with an excellent  AD / DA converter, lies for about two orders of magnitude without reach of other designs.


Of course there is no such sample rate like 1 / 100 fs equivalent to 10.000.000.000.000 Hz - in real world it rather will be 196 kHz - but looking at the scenery from that side gives a good and more intuitive understandable picture of the accuracy involved.
When real world digital audio signals are converted back to analog, exactly above time step accuracy is translated back into the analog signal with the help of the low pass reconstruction filter.


What I want to highlight here is that full 24 Bit audio signal processing is really challenging and to put it simple, you won't get full 24 Bit resolution with less than perfect clocks. Any time you are in doubt about the precision needed for the exact sampling time, remember that for frequencies close to the Nyquist frequency there are only slightly more than two samples for any period from which the original sine wave in exact frequency AND exact amplitude has to be restored.

For those with a philosophic attitude that may ask themselves: "do we hear such small variations in time span like 100 fs ?" - the best answer I found for myself was: "Do we prefer 24 Bit over 16 Bit ?"








Bottom Line




The DeepColor clock proves to be a unique yet beautiful design approach to support digital audio up to the highest resolution.
Everything outlined above is quite easy to understand.
To obtain ultimate precision, balance noise against speed, let SPICE perform some maths, draw your conclusions and listen - that's it.


It may sound unbelievable from today's knowledge about human hearing ability, though to me there seems to be even further room for audible improvement in timing.
But never mind, if you are already happy with what you have - don't worry about all that jitter hype above!
If you are already looking for something better and you'd like to give it a try - you always can ask for PCB boards that will become readily available soon.

If you have suggestions or found a way to do any better I'd be happy, if you dropped me a line.





Keep swingin' !






************************************************************************************************************************






Austria, in January 2007

Grafik 
Michael Gerstgrasser


Updated: 30.3.2007

Grafik