| capslock |
I have been trying to build an ECL oscillator similar do the circuits described in the IQD crystal data book, by Winser and by Unruh (the latter two articles were on the ACG pages when they were still in service - what happenend to them?).
Basically, in this circuit the crystal is used in series resonant mode, it is connected from the non-inverting output to the non-inverting input, which in turn has a 15R load to AC ground. The rest is just for getting the DC setpoints right. The input is shorted to ground by the 15 R resistor for all frequencies except the series resonant mode where the crystal has nothing but its real impedance, which is around 10-20 R for HC-49 case devices.
As the 10216 ECL line receiver is not easily available, I used the single gate. 10H16 PECL line receiver. Apart from being faster (0.5 ns prop delay) which wouldn't hurt in this application, the chip is just the same.
Initially, the circuit would sometimes not start up, sometimes start up at the desired frequency, sometime start up in a spurious oscillation aroung 300-600 MHz. I tried various crystals, load resistors beween 10 and 39 R, loutput pull - downs between 100 and 200 R. No change.
The I found that the returns from the input and output loads had different vias to the ground plane just half an inch apart. I provided another wire bridge. The spurious oscillations were gone, but so was the fundamental oscillation. Even injecting a square wave signal couldn't humor the circuit into oscillation. Any ideas?
Frustated,
Eric |
|
|
| Elso Kwak |
Hi Eric,
First a important question : what is the frequency of the crystal you are using?
From about 20 Mhz and up crystals operate in the third overtone mode and from about 80 MHz in the fifth overtone mode. It also depends on the manufacturer. Especially the overtone crystals are very temperamental to put it mildly.
Personally I don't have experience with these ECL oscillators, but collected some 10116's in the shoebox.:) |
|
|
| capslock |
Hi Elso,
those were all fundamental mode crystals, I tried various specimen from 12 to 20 MHz. All worked nicely when inserted in the classic 74HCU04 circuit.
I followed your link to the Jung-like regulators. I think I have seen something similar in some AD app note, only that they took some great care to avoid oscillation.
Regards,
Eric |
|
|
| capslock |
Redid the circuit on a spare PCB, not having soldered in the ECL-to-TTL translator helped, too.
1) I removed the two low pass filters that are used in the IQD, Winser and Unruh circuits to derive the DC setpoint. Instead, I feed the inverting input directly from the V_BB pin, with a 100 nF to ground. That is the nice thing about using a line receiver gate, it actually provides the DC reference. Not sure that this is critical, but it sure reducec component count by two resistors and two capacitors.
2) Soldered the load resistor across from non-inverting to inverting input instead to ground. Should be the same, as the inverting input is also grounded, but sure removes a couple of mm worth of copper which will have some impedance.
After these alterations, the circuit worked fine with one of the more benign crystals, even with a 39 R input load resistor.
Having implemented these changes into the final board (which has the translator soldered in), it worked most of the time, but sometimes I got the spurios oscillation (150 to 400 MHz). Lowering the input load to 10 R helped. On start-up, the spurious oscillation lingered, but it would always give way to the crystal fundamental mode after a second.
I put in the 16.9344 crystal salvaged from a Sony player. Now I could trigger the fundamental mode only when I played with the scope probe. I lowered the load to 8.7 R - fundamental mode kicked in somewhat more easily. This gave me the clue: the circuit still had too much gain aroung 150 MHz. A 100 pF in parallel to the input load rolled of gain by 3 dB at 150 MHz. Now the fundamental is there right from the start :)
Will post the circuit some day when I get to draw it...
Eric |
|
|
| Elso Kwak |
Hi Eric,
Glad to hear it is working now.
Don't use the three leaded crystal from the Sony. It is more like a crystal <B>filter</B> Use a "normal" 16.9344 Mhz crystal.
The tendency of your circuit to oscillate at a much higher frequency reminds me of my early attemps to build a oscillator with the LT1016. Too much drive on the crystal.
The KWAK-CLOCK uses gentle, minimal drive to the crystal resulting in much lower jitter and absolutely no spurious oscillations.:) |
|
|
| capslock |
Dear Elso,
in the three leaded crystals provided by IQD/C-MAC, the third lead is simply the case. This saves you having to solder a wire from the ground plane to the case.
There are three leaded little blue things, too. These are ceramic resonators which have much higher internal damping than crystals. The Q of a resonator is on the order of 1000 where a 16 MHz crytal in the large HC49 case will have about 80000. The center lead is also ground, but in addition there are two lag capacitors to each active pin of the resonator.
My ECL circuit uses the crystal in series resonance where its impedance becomes equal to its series resistance which is typically about 10-15 R. You want to provide a low impedance drive and load to keep the Q of the circuit. The drive into the crystal is about 900 mV with a square wave. On the load side, I get a beautiful 400 mV sine. In first order approximation, I am dissipating about 20 mW in the crystal. This may be to much as most crystals are specified at 1- 5 mW. That is why I am trying to find the article on the Winser clock implemented into a TEAC that was on the ACG pages. The author (Werdin?) did line out how to measure and calculate the crystal drive. If I don't find it, I'll have to do the math myself...
Your Quak clock appears to use parallel resonance. Here the resonator impedance becomes infinite. To preserve Q, you want to load the circuit as little as possible. The 10 M resistor looks fine. However, I am a little puzzled whether the 1 k source resistor won't introduce too much damping. What do you think? Also, what does the 10 pF in series with the crystal do? I assume it corrects the crystal frequency but this is really a circuit you would expect to find in a series resonant circuit.
What do you think?
Regards,
Eric |
|
|
| ftorres |
Hi Eric,
Found in my archives the paper by Wedin. As I only got a paper version, I copy the whole paragraph.
Here we go :)
"The current through the crystal (and the load resistor) is the square root of the crystal's maximum permissible power dissipation divided by its internal series resistance, I=Pxs/Rxs. With values inserted, 0.5mW and 13 ohms, we get 6.2 mA. The voltage drop accross the crystal is the square root of the maximum permissible power dissipation multiplied by the crystal's internal series resistance, U=Pxs*Rxs. With values inserted we get 0.081 Vrms. The ECL-receiver (0.8Vpp or 0.28Vrms), Url=Uecl-Uxs. With values inserted we get 0.2 Vrms. Finally we get the value of the load resistance as the voltage drop of the load resistor divided by the current through it, Rl=Url/I. With values inserted we get 32 Ohms."
N.B. : 0.5 mW is the value of the max dissipation of Wedin's crystal, and 13 Ohms is the internal series resistance he had get, taking two thirds of the manufacturer's value (20 Ohms)
If you're interested, drop me an e-mail with your address, and I'll send a paper copy of the whole paper.
Elso,
I'm about to test your clock with both the fast comparator and the mosfet we talked about. I have to finish PCB design, but in a week or two, I should have some results to post here.
Cheers |
|
|
| capslock |
Hi Francois,
thanks for copying that paragraph. That will certainly help me. I think I still have a paper copy, but it is still in one of countless not yet unpacked boxes (moved two months ago).
From the literature I have found on oscillators, I guess the subject of how to make a very high Q oscillator has not been delved into. Some authors appear even to be confused whether they use parallel or series resonance.
Where is Limoges? I think I have been there.
Greetings,
Eric |
|
|
| capslock |
Well, I did a few quick calculations. I am dissipating about 2 mW, the crystal is presumably specified for 1 mW.
However, I am not sure that that part is entirely correct. First of all, it assumes everything to be in phase. I guess, in series resonance this is a pretty accurate guess. I will check, though...
However, I am not happy with the treatment of the dissipation. After all, the crystal is driven by a square wave (at least with a fast ECL unit), and it outputs a sine. I thought he commented on that, I think. Do you recall what the article was called?
I am beginning to think about some improvements. For example, the resistive load might be replaced by an LC series resonant circuit. Also, the ECL line receiver could be built in a discrete circuit. This way, it would be possible to limit the drive and still maintain a low impedance load.
Greetings,
Eric |
|
|
| ftorres |
Hi back,
I dunno if you already know it, but a lot of interesting informations can be found at
http://www.telequarz.de/info/info_q.html
(this is the direct link, but can be reached in a regular way to see all other docs : http://www.telequarz.de/index_e.html )
And, lucky guy, there's a nice book on quartz and oscillators, and IT'S IN GERMAN :( !!
Most instersting chapters (IMHO, and if I may guess from my poooooooor german) are chapter 2 (mostly theoretical) and chapter 6, with oscillators analysis.
Chapter 2 is summed up in the first pdf than can be downloaded.
Wedin's article is called "Re-clocking TEAC VRDS-T1 and TEAC VRDS-7"
And, oh, Limoges is 300km south from Paris, near the center of France ;) |
|
|
| capslock |
Thanks, this is a treasure-trove. Let me know if there is anything you would like to have translated.
Eric |
|
|
| Elso Kwak |
Hi Francois,
Just tried the BF980 Dual gate MOSFET in place of the AD8561 comparator in the KWAK-CLOCK as in the Elektor circuit kindly provided by you.
The good news is it works!; the bad news is that the <B>magic is gone!</B>. Sonically it is a big step backwards. I suspect the MOSFET produces a clipped sinewave at best and not a square wave as desired.
Gone is the incredible depth of the soundstage and the fine bass. Also found a lack of definition in the sound with the MOSFET. |
|
|
| Elso Kwak |
Hi Eric,
I only wanted to warn you as the three leaded Sony crystal produces a much worse sound than the two leaded type they also use in some of there players. Notably the types with R2R ladder type dac's. I am aware that the third leg is connected to ground. Just <B><I>practical</B></I> experience sharing with you.
As for the component value's of the KWAK-CLOCK; these are optimised by ear and a close look at other circuits.
I also watched for startup without problems. The KWAK-CLOCK circuit accomodates a wide range of crystal frequencies from 8 to 20 MHz. I have tried a active current source in place of the source resistor. It was NOT a improvement. I have tried <B>MANY</B> other things, including bipolar transistors and other FET's, too much to elaborate on here. |
|
|
| ftorres |
Elso,
Good and bad news :rolleyes:
Some care must be taken with the dual gate mosfet. First, there MUST be a capacitor (1nF or less) between your oscillator and G1 of the mosfet (I don't have your schametics handy, and I don't remember you have one). Secondly, you will probably be better with a drain resistance of 560 or 470 Ohms for the mosfet. And last, I'm not sure that G2 should be tight to 5V. I was planning to use a 100K trimpot to vary G2 potential. This voltage acts as a kind of commutation threshold for the mosfet. I will compare with yours using a 300MHz scope, maybe it will help to understand. Last of the last, I will check the BF980 datasheet and compare it with the BF998's, which I'm planning to use, to see if their caracteristics are comparable.
Eric,
Thanxxxx for your translation proposal. I'll check the interesting points for me, and ring you back :) |
|
|
| capslock |
Elso,
I am beginning to see what your circuit which is basically a Colpitts oscillator does. The crystal is used at slightly above the series resonant frequence where it appears inductive. This inductance will form a parallel resonant circuit with the series combination of the 39 and 68 pF capacitors. The crystal appears inductive only for a narrow frequency region, and somewhere in this region the parallel circuit has its resonance, i.e. its impedance will become high, at roughly 4x times ESR of the crystal. Hencce, at this frequency the input short is almost removed.
The transistor source drives this parallel circuit through the capacitive tap provided by the two capacitors. The drive impedance should be as low as possible, and it is derived from the dynamic output impedance of the transistor in parallel with the source resistor. That is why a current source did not do any good.
I don't know what the operating point (is it a JFET or MOSFET?) is in your case but I assume the dynamic output impedance to be on the order of 200-400 R. This is a fairly low output impedance but I think there is room for improvement. You could buffer the source by a NPN emitter follower and feed the emitter voltage back to the parallel resonant circuit. You would have the same AC output and drive voltage (with a 0.6 V DC offset) but you can easily achieve <5 R output impedance by running the NPN at 5 mA.
The 10 M gate bias resistor should not really degrade Q compared to the < 500 R drive impedance. However, I am not quite sure whether it is really in AC parallel with the drive. If it is not, it might be degrading Q. You might want to use an AC bootstrap circuit to increase the impedance at the gate and listen...
Eric |
|
|
| capslock |
Francois,
it is actually chapter 7 that deals with phase noise. Unfortunately, all the greek symbols are missing, so it is a bit difficult to read. Most of it is about definitions and measurements of noise.
But there are also some guidelines for a low phase noise circuit (as in any cookbook):
- don't degrade Q (naturally)
- run the quartz at a high dissipation (this will degrade long term frequency stability, but that does not bother us...)
- if using a bipolar transistor, choose a low frequency, low noise type with high DC current gain h_FE and low base resistance (reason given: phase noise is at several Hz around the fundamental, and here the LF noise performance is more important than the HF noise performance)
- choose a transit frequency of about 5x the crystal frequency (this I don't understand, because then the AC current gain will be roughly 5 and hence there will be a significant load on the crystal, roughly 5x R_E for an emitter follower)
- PNPs are less noisy than NPNs (one never realy is finished with learning...)
- JFETS are less noise than bipolar transistors
- MOSFETs are noisier than bipolars
- GaAs-Transistors are low noise at HF but have very high noise at LF
- make sure the amplifier remains in its linear region (usually amplifier non-linearities are used to limit the impedance but this will cause side band noise)
- implement the amplitude control seperately from the HF amplification
- overtone crystals usually have higher Q when implemented into a real circuit
I guess I will try to compute the effective Q of various circuits (Colpitts, Pierce, ECL) to determine when a resistance is desirable and when it is not....
Eric |
|
|
| ergo |
To capslock
The creek symbol problem has to be a problem with your acrobat reader. I can see these just fine. I'm using Acrobat Reader 4.05c I tried version 5 for a while but it seemed to have many problems I don't have with older version (schematics getting fuzzy when zooming etc.) So if you can try the version 4...
Ergo |
|
|
| ftorres |
Eric,
Thanks for this insightful summary ! All the points you mention confirm what I have read before. Concerning the transistor's transition frequency, it puzzles me too. Take a look at http://sss-mag.com/pdf/lonsosc.pdf , they have a different approach ;) Or maybe they don't speak of the same thing. The more I try to learn, the less I know :(
Cheers |
|
|
| capslock |
As promised a long time ago, my oscillator circuit!
constructional notes:
- The crystal feeds the output back to the input, but this signal gets shunted by R1 and C1. On resonance, the impedance of the crystal is minimal. Drive to the crystal has very low impedance, but the shunt impedance is also low. So I am not sure about the claims repeated throughout the literature that this or similar circuits actually have a very high Q. How does one compute Q here?
- Connecting the shunt across the inputs was the only way to make the circuit work, and it is probably the best way, too.
- The circuit worked fine with R1 = 9 R, however, I was not sure if I wasn't overdriving the crystal. Doing the math for the power dissipation isn't trivial because the drive voltage is not a sinusoidal as suggested by the IQD app note or Wedin's article. Also, without a model for the particular crystal, one can never be sure where exactly (slightly of series resonance) it really resonates.
High drive is good for jitter performance but will degrade long term stability. Also, I am not sure that a severely overdriven crystal will still give good jitter performance.
- Make sure not to use a crystal in a HC-49S or similar small case. The crystals in the big cases can stand higher drive and they have lower series resistance, i.e. higher inherent Q.
- C1 was necessary to prevent the circuit from starting up at about 700 MHz occasionally. Required value depends on layout. Make sure that 1/(2*PI*f_crystal) is still at least 5x larger than R1.
- C2 is still connected to ground in my prototype. The next PCB will have it connected to V_CC which is the internal reference node of the ECL line receiver.
Next installment: possible improvements
Comments? Questions?
The attached file is GIF even if the extension is zip. Why is this forum so picky about extensions?
Greetings,
Eric |
|
|
| capslock |
This attachment shows a discrete version of the ECL line receiver. D-bar should be connected to V_BB which should be -1.3V relative to V_CC and may be generated by a simple divider.
Using AF low noise transistors should result in lower low frequency jitter. I have not yet built the circuit, so I am not sure if the circuit is slow enough so the output will be sinusoidal. One could include C1 to smear the edges. This should not degrade the output impedance which depends mainly on the current of the emitter followers.
One basic problem remains. While the differential amp is essentially a low noise and reasonably linear amplifier, some nonlinearity is needed to keep the oscillation amplitude in check. The collector resistors are chosen at 220 R to limit the swing before any of the transistors goes into saturation (that was the original idea behind ECL). In our case, it will also limit the drive voltage to the crystal, so the drive to the crytal can never be spectrally pure. C1 may filter out some of the harmonics, but the rest is left to the crystal to handle.
One could make the diff amp even more linear by using a higher supply voltage and emitter degeneration resistors. But when compression sets in later, the circuit will automatically settle at a higher oscillation amplitude. Therefore, some sort of AGC (automatic gain control) is needed. One could replace the 750 R tail resistor by a voltage controlled current source. My problem now is that not being a radio nut, I don't know how to rectify a 17 MHz signal decently. I have seen one AGC circuit that I'll post later but I am not sure if it is going to work well.
Greetings,
Eric |
|
|
| capslock |
This is from D. Nührmann, Industrielle Schaltungstechnik (again gif disguised as zip). The output signal is AC-coupled via the 2.2 uF capacitor to the AA143 germanium diode. It will clamp the negative excursions, while the positive excursions charge the 220 uF + 12 nF capacitors through the 39 k resistor. The op amp will control the tail current in trying to balance its inputs. The amplitude is set by controlling the reference voltage on the non-inverting input.
Is the AA143 adequate for HF clamping? Is there a more suitable replacement, i.e. a HF Schottky diode? Is there a simpler AGC circuit?
Thanks,
Eric |
|
|
| Jocko Homo |
Been watching this thread casually, and I am planning to work on a schematic for a low-noise oscillator that uses NPN transistors, so it will be easier to build.
Recommendations on program to use that will let me post a readable schematic? (I never use schematic capture........too old to care about learning something that new.) [joke]
Jocko |
|
|
| ergo |
Jocko, I find MicroCap to be absolutely wonderful for drawing schematics. It a very powerful simulator etc. but also very suitable for schematic drawing. There is a demo version available, with no time limitation and 50 part simulation restriction, but there is no restriction in just schematic drawing.
I have tried many such programs and MicroCap is in my opinion by far the easiest and most intuitive to use as circuit drawing tool. There are also nice tutorial scripts that get you running in no time :)
http://www.spectrum-soft.com
Check the JLH amp thread for sample schematic made with MicroCap
http://www.diyaudio.com/forums/show...15&pagenumber=4
UHH. Hope you forgive me for such preaching :)
Ergo |
|
|
| Jocko Homo |
Preach on.
Let's see if I can upload this kludge schematic made a different way.
Nope too big. I have a .gif of it if anyone is interested. Email me.
Jocko |
|
|
| Jocko Homo |
OK......now it is my turn to be the idiot..........
How do you save this in a format that will be accepted here?
Jocko |
|
|
| Jocko Homo |
Let's try it again.......
This originally appeared in EDN in 5/73. I have seen it several places, including the ARRL Handbook.
The trick about this is that the crystal acts as both a low-pass, and sideband filter. This version is 11.2896 MHz. C4 should be about 680 pF, but I increased it to keep the harmonics down in Q3. Probably could have done the same by playing with R11, but I just grabbed the handiest values around. That way anyone should be able to build it.
I did not include the power supply decoupling capacitors. I leave that part up to your discretion. |
|
|
| wildmonkeysects |
Hey all:
In wireless/telecom circles where I wear my golden handcuffs, the Colpitts topology is widely regarded as the most straightforward low phase noise osc: if you want clean, do a differential jfet colpitts and feed it to a high speed comparitor is the word I have been hearing. Having ignored a number of subtle hints, and finally listening to Elso, I heated up the slobbering iron, and dove in a built a couple. So far, wow. Night and day sonic improvement over stock CMOS and comparitor oscs replaced.
There are still issues, largely implementation. The comparitor does not swing high enough with a 5 volt supply, so it has a 6.25 volt supply to swing 5 volts. When driving even a single CMOS/TTL load, the comparitor "causes" sufficient ground bounce to create glitches in the Colpitts sine and sine bar outs, jfet sources. Mostly common moded out, but I still don't like them. Yes, even with rf scope probes. Version three will have a split ground plane. Maybe it will be ready for primetime.
Sorry for the fuzzy sch, had to squeeze it to fit. Source resistors are 475r, gate resistors are 10M, comp snubber resistors are 100r, top cap is 47p, bottom cap is 100p.
Not in sch are bypass caps galore. Jfet supply is an LT1021, comp supply is an LT1086. Built "3D" on a copper foil ground plane. XTAL is sorbothane mounted, as is whole assembly. Caps are silver-mica, jfets are 2n5457s for now, matched Idss of 4 mA, comp is MAX1116 for now. Jfets draw ~~2mA ea in circuit, source voltages ~~0.8v.
Enjoy...
WMS |
|
|
| capslock |
Jocko,
could you elaborate on how your comment that the crystal works as both low pass as well as side band filter? I seem to have trouble understanding circuits that were divised in the 70s when engineers were still thinking in tubes...
Does Q1 act as a free-running oscillator?
Puzzled,
Eric |
|
|
| Jocko Homo |
Problem with JFETs is that they don't have enough transconductance to really drive the crystal hard. Diff pair might get the drive a bit higher.
My telecom days are over. At least for now. Which also means I do not have access to MY H-P phase noise setup.
Ther signal is picked off from the 22 ohm resistor. It has to go through the crystal to get there. Neat idea. Wish I had thought it up.
Jocko |
|
|
| capslock |
Anybody have a foolproof way to measure phase noise? I have a good number of high bandwidth real time sampling scopes around but I have the feeling I cannot trust their clocks and converter aperture jitter....
Eric |
|
|
| wildmonkeysects |
Yikes, the next attachment will be less fuzzy...looks like a lot of monolithic opamps sound to me...
Hey, this board is very refreshing, seems very civilized ... an open sharing of ideas with a minimum of flames and egos.
Looking forward to posts of sonic results with various osc topologies.
Oh, yes...jfets are also known to suck in terms of 1/f noise, but whatever spectra they exude, seem to work well as a Colpitts. The idea is that one does not need a lot of gain, just enough to overcome entropy and keep the tank ringing. It is normal to wonder WTF when first looking at a Colpitts! The differential arrangement relies on intentional injection locking for once, which happens out of phase. The signal level could be higher if one biases the gates above ground, but that would introduce power supply coupling which in my book is evil.
Hoo boy, measuring phase noise is an art unto itself. Spectrum analyzers with low noise options are the best bet for such freq domain measurements. In the time domain, fast scopes can only be of marginal use. One chip house I know of uses old fashioned analog storage scopes and, get this, a magnifying glass, but is limited to levels around a couple hundred pS of jitter.
As Elso et al have gently pointed out, the ears are a wonderful instrument. The tools won't specifically lie to you, just won't tell the you the whole truth. I can hear deliberately induced thermal tails and supply induced grunge quite easily with familiar music on a familiar system, but trying to measure them is like splitting hairs. I'd rather grok what the circuit is doing and optimize by ear, but that is just me...
Enjoy,
WMS |
|
|
| Jocko Homo |
Drew it a different way........forgot C8 too.
If you short R4/C4, and connect the left side of C5 to the dashed lines (which don't show up now!), it looks like any other oscillator.
The only way is with a phase noise setup. I guess you good build one with if you already had a low-noise oscillator, mixer, and spectrum analyzer.
Jocko |
|
|
| ftorres |
Eric,
Nice design with ECL receivers, and it reminds me of a sch. that can be found on http://www.euroquartz.co.uk/pdf/application-notes.pdf (page 2). I'm no sure you really need the cap 'cross the inputs, but it is safer. Maybe your parasitic oscillations are related to the "real" circuits stray inductances and capacitances. Have you tested it with SMT parts, including the crystal itself ?
Concerning oscillators, I've been busy toying with various designs to plug into my CD player. Most of all led to improvements, but the best ones ('til now ;) ) were Elso's KWAK/Colpitts clock and a schemo I borrowed from an old Elektor magazine (common base Butelr XO). On both designs, I've tested two output stages, one with a fast comparator (AD8561) and an ECL stage (MC10ELT21) as per as AD's application note AN419 ( http://www.analog.com/library/appli...ators/an419.pdf )
(Yes, Elso, I've given up with the dual-gate mosfet buffer - you were right, as usual ;) )
On a scope, all the designs had quite the same characteristics (rise time, levels, etc...), but the ultimate musical tests gave surprising results :
- Elso's clock was slightly better with AD8561 output stage
- Butler common base was better with ECL stage...
And finally, I was unable to hear a difference between the two designs (each one with the right output stage) ... And both clocks make wonderful music, with a lot of space, great bass, full of air and details.
For the interested ones, I've attached the schematics of the Common Base Butler XO. It needs a series resonant Xtal, but frequency can be slightly trimmed with the variable cap (which can be removed if needed). The left schemo is the basic (but working) one, and the right one is improved a little bit with diode clamping and a much better insensivity to power supply variations, but at the cost of a larger part count. As far as I have tested, you don't have to change any component value if you change the Xtal frequency.
Any comments/improvements welcome.
Greetings, |
|
|
| ftorres |
Hi again :rolleyes:
I forgot in my previous post to mention some experiments (Maybe off-topic here) I made with high frequency oscillators (96-98 MHz) for asynchronous reclocking in DACs (see http://www.diyaudio.com/forums/show...=&threadid=1594 for a start point) . To sum up, the idea is to use D flipflops just before the DAC chips to reclock the digital stream using an asynchronous high frequency clock. I'm completely unable to technically explain what happens to the sound when doing so, but I wanted to test it.
To jump to the conclusion, I don't know how it does, but it works ! The improvement gap is not as large as when I first plugged a high quality clock in my CD player, but the improvement is here. How to qualify it ? IMHO, it reduces harshness, making the sound more "analog". But it's only one day old , and I haven't had enough time to perform careful listening tests... more to come...
But back to the off-topic topic :p of this thread. The most challenging part (for me) of this little project was to find a high quality 96-100MHz clock to feed the FFs. I've tested various designs I found in litterature : Analog Devices' AN-419 Butler XO, Butler common base C-Tap, Butler Emitter Follower, Pierce, Harmonic ECL... If you're interested, I can post schematics/links.
All these designs were buffered by an AD8611 ultra fast comparator or by the now-classical MC10ELT21 ECL IC. I finally settled on the ECL IC, which gave better results, as well on the scope as in my ears.
And last but not least, I've attached the schematics of the XO I found to be the best IMHO, in terms of stability, ease of startup, simplicity...and sound (for the moment). It's called the emitter coupled harmonic XO. May be the three adjustement points can frighten you, but it's very tolerant, and once trimmed, it will always start.
Feel free to test and comment it. |
|
|
| tvi |
Wenzel Associates Inc have a page on "Techniques for Measuring Phase Noise"
Also thought I'd add the high-end clock from elektor from a few years ago.
Schematic was done with Proteus-lite from <a href="http://www.labcenter.co.uk/">www.labcenter.co.uk</a>, free with some limits but very useable.
Regards
James |
|
|
| Elso Kwak |
Hi Wildmonkeysects,
I builded your modification with two FET's. Strictly speaking this is no longer a Colpitts oscillator. However I adhered to the original parts and values of the KWAK-CLOCK i.e. J309 and AD8561. The crystal was "hung" by two 10pF caps between the gates of the FET's. I did not have the problems you encountered; it worked right away. Sonically the result was <B>exactly</B> the same as the original KWAK-CLOCK.
I also builded a version of the KWAK-CLOCK with just a J309 FET added with 1M gate biasing ad 1k source resistor. Source connected to the - input of the comparator. Again this works but sonically no difference to the original circuit.
You are using a different FET and LT1116 comparator [probably misspelled as MAX1116 which is a 8 bit DAC]. As I experienced it does make a big difference which FET and comparator are being used; I strongly encourage you to get a J309 and AD8561 as these produce the best results. I have been using the 2SK117BL and LT1016 with worse results.
From Rudolf Broertjes I got the suggestion to add a large 6800µF capacitor across the clock supply. Though very sceptical about this mod I tried a 10000µF electrolytic I had laying around. This mod produced a slightly better bass and slightly more open soundfield. I don't have a explainnation for this result; lower noise on the supply? I am using a LT1086-5 for the clocksupply.
Also builded various supplies like LT1021-5 reference and OP27 opamp in a Walt Jung like regulator, BC550C/TL431 series regulator with PI-filter in front of it. The latter produced a soundpicture slightly more "at ease". I did not like the Jung regulator in this application and still prefer the LT1086-5 or the LT1086; with bypassed adjustment pin; used for the supply of the CDP; in my setup.:) |
|
|
| Jocko Homo |
2SK117BL? Are my eyes playing tricks on me again?
I need to find the schematic of this thing........
Jocko |
|
|
| Elso Kwak |
Hi Jocko,
Stirring things up again......?!
Schematic sent!
I choose the 2SK117BL initially because it was laying around and is very low noise. People keep reminding me noise= phasenoise=jitter.
I am not so sure of this simple relationship.:) ;) |
|
|
| capslock |
There are a lot of cookbook recipes like preserve Q, use FETs, use low noise AF transistors, use very sinusoidal drive, ...
However, when it comes to optimizing a given circuit or deciding whether one circuit is better than another, it is not clear which of these points will be the most important in a given application.
I particular, it seems to be very unclear of what the effective Q of a crystal inside a circuit is.
Might this be a way to simulate Q: simulate the oscillator, inject some sinusoidal of various frequencies close to the fundamental into on lead of the crystal using a high-value resistor and see how much of this signal is rejected by the circuit?
Puzzled,
Eric |
|
|
| Jocko Homo |
Eric:
Just got around to fully reading your post of 2/28.
The reason you don't want an active device with too high Ft is that there will be more parasitics. So chose the slowest one that is fast enough.
JFETs do have lower noise, which is good, but they ususally don't have enough gain to drive the crystal hard enough. I've built RF circuits with audio transistors, many times, and have always gone back to the RF types. I needed the gain.
Power supply noise is probably a bigger contributor of phase noise than the active device.
Somewhere I have a good book on oscillators. I seem to recall the author feeling that using JFETs for the lower Q loading was overrated.
Jocko |
|
|
| Elso Kwak |
Hi Jocko,
Of all the bipolar transistors and FET's tried the J309 was giving the best <B>SOUNDING</B> result.
I tried BC550C; 2N4401 (best of all bipolars); 2N2369A; half a MAT02; 2SK117BL; 2SK161GR (giving problems), J309; J310 (works as well); U440 in the Wildmonkeysects circuit (did not work)
My way of working may strike you. Being a chemist by education I find the PRACTICAL result i.e. the sound the most important issue.
The KWAK-CLOCK is the result of my experiments using my ears as the ultimate judgement.
Same story for the comparator. I started with the LT1016; AD9696 and switched to AD8561.
Did you know that LCAudio uses a 5 GHz transistor? Speaking of overkill.;)
<I>BTW the 0.01 µF coupling cap between FET and comparator and the 100k resistor are superfluous. Omit these components and directly connect the source of the FET with the positive input pin#2 of the comparator (no change in sound though)</I>Helas.:) |
|
|
| capslock |
Hi Elso,
the reason for the observed ("obheard"?) differences might be that a comparator will work with a very small sinusoidal input signal whereas an ECL input needs a couple of 100 mV for a full swing. Also, even a PECL devices primary reference for all input signals is not ground but the positive supply.
Apart from the old Moterola ECL data books and app notes which are still available as PDFs at OnSemi, Tietze/Schenk, "Halbleiter-Schaltungstechnik" is very instructive. The 11th edition has been expanded also to give much more information on how to design current sources and op-amps, both in discrete and integrated design. I haven't found this much concentrated information anywhere else yet.
Greetings,
Eric |
|
|
| Elso Kwak |
Thanks Eric,
I did not try the ECL device yet. Francois did.
Thanks for the reading suggestion, I think I am beyond that.......;) |
|
|
| Jocko Homo |
Just so you guys won't think I am closed-minded about evertything, I tried the old standby......2SC2240.....(sonnya, are you reading this? Heh-heh.) in the oscillator I posted.
Same voltage output. So, what next........
Build two more, one with '918, one with '2240. Get a mixer out of the old junk box, and use one with a '2240 as the reference oscillator. Compare the downconverted spectrum on my FFT (when I get it back that is: on loan.) Then we can come up with a relative phase noise measurement.
Standby for results, just don't hold your breath.
Jocko |
|
|
| capslock |
Sorry Elso,
I should have reread the last couple of threads. Somehow, I was sure you observed the sonic differences between ECL convertor and comparator, but I was wrong!
I have been doing analog electronics for ages, and I had digested the previous editions of Tietze/Schenk thoroughly. Still, I found many things I didn't know or had been trying to find for a long time in the new edition. The reason is that is has been expanded to cover integrated analog circuit design. Most of the things can be applied to discrete circuits, too, if you throw in a couple of degeneration resistors to deal with transistor matching problems.
Groetjes,
Eric |
|
|
| dimitri |
>Then we can come up with a relative phase noise measurement.
>Standby for results, just don't hold your breath.
Dear Giacomo, how do you feel, two years is enough for standby? ;) |
|
|
|
