As I stared at the Taitron data I began to recall some of the details when I used 1N5231s (nominal 5.1V) as zero-tempco references, interested primarily in low tempco. I would experimentally adjust the current to determine the operating point, and I recall some parts around 4.78V. But these currents may not have corresponded to the lowest noise, which probably skewed my conclusions. Motchenbacher and Connelly affirm that avalanche diodes are per se noisier, although they add that most of the avalanche noise is amenable to filtering (e.g., not associated with excess low frequency noise).
So DVV's experience may well be typical. Time for some modern measurements (mine date back mostly to astronomical instrumentation days, circa 1976). I see Taitron have some fairly local distribution.
Quite so, Brad, it seems we both could do with a refresher course, things have improved since the good, old days, it's a whole new world now.
And not everythings is as it is touted to be. A while ago, I found some Chinese capacitors which turned out to be rather good, certainly better than expected, and giving the big guns a run for their money. To be sure, not as good as the established lines from Nichicon, Elna, Sanyo at al., but better than the average, so to speak. At very good prices. Not so 10 years ago, time to investigate newer parts.
On low noise transistors - my favorites are Toshiba's 2SC2240/2SA970. The problem is getting the real deal rather than some rebadged duds, but we all have that problem.
And not everythings is as it is touted to be. A while ago, I found some Chinese capacitors which turned out to be rather good, certainly better than expected, and giving the big guns a run for their money. To be sure, not as good as the established lines from Nichicon, Elna, Sanyo at al., but better than the average, so to speak. At very good prices. Not so 10 years ago, time to investigate newer parts.
On low noise transistors - my favorites are Toshiba's 2SC2240/2SA970. The problem is getting the real deal rather than some rebadged duds, but we all have that problem.
One of them is dead. Richard Hay of Nytech. Of all of the Gang of Four he was the really interesting person to talk to. He was even more in favour of minimal design than me, what he didn't know wouldn't fill a very large book when our subject. He knew more about distortion and how to measure it than anyone I have met. He showed me a way and I don't think I will share it. I have never seen it written about and is so simple. I have hinted at it many times and no one has picked up the ball. He came from Radford. I must say some of the Radford designs were ugly sounding. This is strange as the ZD50 looks to be exactly what I would think up. I must build a ZD50 because it should be out of this world good. Why it wasn't beats me. It has no obvious problems. ZD means zero distortion as at the time it could not be measured.
Flat Earth was a very correct thing. If it doesn't sound better it isn't. It is for the engineer to find out why. The public buy on sound alone. In the long term this is totally true. What people here think is just a bit of fun ( at best ). The truth is the reality of what is sold. Like good food people do not mind having something good. The NAD 3020 starts to be a good thing. Not quite my cup of tea. I always get a bit interesed to listen if someone has one. It has no obvious problem except being a bit dull.
As far as I know no one in the Flat Earth was interested in LED's. Naim made no special reference to them or cures. No idea who started the rumour. Doubless most people filtered them as it was worth 10 pence to say " yes we do understand and do prevent any risk".
The problem is the flat earth stuff was usually dreadfully flawed but sold people hook line and sinker into the whole foot tappy PRAT crap(tm) that a whole generation lost the ability to think or listen for themselves. I have very little brit kit. The STA-25 mk3, DPA50S, roksan and SME arm. Until I can afford some of Tims stuff or an SME turntable nothing else really interests me.
OK maybe a DNM setup for the study, but only cos Dennis had a caterham.
OK maybe a DNM setup for the study, but only cos Dennis had a caterham.
DNM. Oh yes I would buy that. Dennis asked me how much current would QED 79 take. I said 500 A and Dennis " At last someone who knows ". 27 Amps conservative, 500 A if you had to would be OK. Dennis was talking transients. I use a generic of DNM cable.
Ray Collins lent a certain man from Scotland a book called BS Baffles Brains. It was £30 when £30 was a lot of money. Bascially what it said was. You are the expert and however expert they are they are not as much of an expert as you about your product. Invent words that sound right ( timing ) . If the person you are talikng to repeats your word you must shut up. You have won the arguement. Not only that, the person becomes your salesman. The next thing the book said is highly intelligent people are usually very slow thinkers. If you have equal intelligence you have the advantage. If the speed at which you seem to think is fast they will take you to be more intelligent. They are also super arrogant and will not easilly understand how they have been upstaged. They especailly are un-use to loosing. They become the company's evangellists as brains is the one thing they admire.
All selling is done this way more or less. Inventing words is the hook.
The telling of the story must be true if it is to have any sucess. The story that is not told is the lie that we know but can not easilly put our finger on. My favourite quote " perfect sound that lasts for ever is getting better". Not sure if quoting myself ? How Philips sold CD was a total lie. They knew it had toubles. Linn never told me a lie as best I know, I strongely suspect the reverse. They may have held strong views. Philips needed the money back. One can sort of let them off by saying it had the potential. Linn perhaps were quoted too often by people who should have looked more widely for other products.
Ray Collins was the rep for Ariston. The other gentleman his friend from another company. Ray would never say a bad word against him. He said his friend was always listening to things night and day. That makes a big difference.
Ray Collins lent a certain man from Scotland a book called BS Baffles Brains. It was £30 when £30 was a lot of money. Bascially what it said was. You are the expert and however expert they are they are not as much of an expert as you about your product. Invent words that sound right ( timing ) . If the person you are talikng to repeats your word you must shut up. You have won the arguement. Not only that, the person becomes your salesman. The next thing the book said is highly intelligent people are usually very slow thinkers. If you have equal intelligence you have the advantage. If the speed at which you seem to think is fast they will take you to be more intelligent. They are also super arrogant and will not easilly understand how they have been upstaged. They especailly are un-use to loosing. They become the company's evangellists as brains is the one thing they admire.
All selling is done this way more or less. Inventing words is the hook.
The telling of the story must be true if it is to have any sucess. The story that is not told is the lie that we know but can not easilly put our finger on. My favourite quote " perfect sound that lasts for ever is getting better". Not sure if quoting myself ? How Philips sold CD was a total lie. They knew it had toubles. Linn never told me a lie as best I know, I strongely suspect the reverse. They may have held strong views. Philips needed the money back. One can sort of let them off by saying it had the potential. Linn perhaps were quoted too often by people who should have looked more widely for other products.
Ray Collins was the rep for Ariston. The other gentleman his friend from another company. Ray would never say a bad word against him. He said his friend was always listening to things night and day. That makes a big difference.
zeners measured
A 5.1V 1N5231B 0.5W part with 10mA has a noise voltage spectral density of about 11.3nV/sq rt Hz, that is, 3.2uV rms in an 80kHz passband. This is without any capacitor in parallel.
A 22V 1N5251B 0.5W part with 11.5mA has a noise voltage spectral density of 193nV/sq rt Hz, that is, 54.6uV rms in an 80kHz passband. So it is appreciably noisier even accounting for the higher voltage.
If the current is reduced to the 250uA on the Taitron part specs, for the 1N5251B the noise soars to 6.4uV/sq rt Hz, 1.82mV rms in an 80kHz passband.
For the 1N5231B and 250uA, we get 19.9nV/sq rt Hz, 5.6uV rms in the 80kHz passband. This is well below the rated current and the d.c. voltage is only about 4.1V.
So these are not too bad, especially for the 1N5231B. In both cases the 80kHz passband makes the low frequency excess noise a relatively small contribution.
I measured a couple of zeners, and although the Taitron data is for the rather low current of 250uA for any of the listed parts, and the noise density is the maximum and not the typical, the results do suggest that the Taitron data is probably pessimistic.
A 5.1V 1N5231B 0.5W part with 10mA has a noise voltage spectral density of about 11.3nV/sq rt Hz, that is, 3.2uV rms in an 80kHz passband. This is without any capacitor in parallel.
A 22V 1N5251B 0.5W part with 11.5mA has a noise voltage spectral density of 193nV/sq rt Hz, that is, 54.6uV rms in an 80kHz passband. So it is appreciably noisier even accounting for the higher voltage.
If the current is reduced to the 250uA on the Taitron part specs, for the 1N5251B the noise soars to 6.4uV/sq rt Hz, 1.82mV rms in an 80kHz passband.
For the 1N5231B and 250uA, we get 19.9nV/sq rt Hz, 5.6uV rms in the 80kHz passband. This is well below the rated current and the d.c. voltage is only about 4.1V.
So these are not too bad, especially for the 1N5231B. In both cases the 80kHz passband makes the low frequency excess noise a relatively small contribution.
I seem to remember a string of 1N4148 was OK. How about the double transistor CCS? Dejan has a point about as high a voltage as possible. Might Early effect be about CCS also. I made a LM317 current sink for a valve circuit. A BD139 with 2 x 1N4148 had more distortion, 2 nd harmonic. 3 x 1N4148 did not.
I recently helped someone design a small speaker. More in the Linn Kan mold than LS3/5A. A 3/5A version was made and was not liked. Every time we removed a bump we lost some magic. The speaker is basically two first order filters with a strange twist. We have a 24 kHz first order filter also. The tweeter is flat to 45 kHz and will have output well past 50 kHz. For my troubles the prototype has returned to me, Colleen was hoping this might happen. An answer to her space problems. On making the crossover tidy I had forgotten the 24 kHz filter. So I listened without it. Very nice but a bit too much. I then put it back in. Wow it sounds beautiful. I am surprised an old guy like me can so easilly hear that. Before anoyone says I am not clainming to hear 24 kHz of even 18 kHz . This difference is totally unexspected. It is one I had suggested but not tried if memory is right?
Then I remembred. The Quad ESL 57 rolls off at 18 kHz. It could be much higher if Quad had wanted it. ESL 57 sounds a bit shut in. I would liken what I hear on my little monitors as very like the Quads only not so obvious it is filtered. The sound is like a better version of the Gold Audax units. What people overlook is filtering a fast unit retains the advantage that speed brings. That is better than high mass not able to move.
My ultra cheap Monacor DT74/8 tweeter is like a poormans version of that ribbon tweeter. It has the very high output also. If I were to put the ribbon in my OB speakers I would need amplifer to get the output. Having heard the little speakers I would like to do that. It has taken me 58 years to build a speaker, now I do two. I repaired them for > 23 years so guess that was the training period. Heybrook HB1 was a nice device. Hardboard with copper nails for the crossover. Piece of paper as a nail guild and as quick as PCB to do.
Then I remembred. The Quad ESL 57 rolls off at 18 kHz. It could be much higher if Quad had wanted it. ESL 57 sounds a bit shut in. I would liken what I hear on my little monitors as very like the Quads only not so obvious it is filtered. The sound is like a better version of the Gold Audax units. What people overlook is filtering a fast unit retains the advantage that speed brings. That is better than high mass not able to move.
My ultra cheap Monacor DT74/8 tweeter is like a poormans version of that ribbon tweeter. It has the very high output also. If I were to put the ribbon in my OB speakers I would need amplifer to get the output. Having heard the little speakers I would like to do that. It has taken me 58 years to build a speaker, now I do two. I repaired them for > 23 years so guess that was the training period. Heybrook HB1 was a nice device. Hardboard with copper nails for the crossover. Piece of paper as a nail guild and as quick as PCB to do.
I get technically sabotaged by my wife and have no Internet for 6 days, and you guys lose steam?
On the 16th, my old style copper cable telephone line went suddenly dead. The ost office service man came and stated that something cut the sable, most probably my wife as she was nailing her 50+ paintings on our walls, after we just finshed redecorating the place. I am now on a newly acquired cable phone and Internet line, which completely bypasses the old telephone lines, as well as their limitations, since the system cannot go over 10 Mb due to line bottlenecks, while the new system does 60 Mb (so I am told).
So I was off the hook, so to speak, for 6 days and you guy just faded out.
Brad, the repose gave me time to think things over a bit. On the one hand, we do not like or want the inevitable zener noise, having agreed that transistors as CCS work better with higher voltages. I don't trust modern LEDs because their properties tend to vayr with manufacturer and series. This leaves me with only standard diodes, like 1N4148, with 3 or 4 in series Agreed?
On the 16th, my old style copper cable telephone line went suddenly dead. The ost office service man came and stated that something cut the sable, most probably my wife as she was nailing her 50+ paintings on our walls, after we just finshed redecorating the place. I am now on a newly acquired cable phone and Internet line, which completely bypasses the old telephone lines, as well as their limitations, since the system cannot go over 10 Mb due to line bottlenecks, while the new system does 60 Mb (so I am told).
So I was off the hook, so to speak, for 6 days and you guy just faded out.
Brad, the repose gave me time to think things over a bit. On the one hand, we do not like or want the inevitable zener noise, having agreed that transistors as CCS work better with higher voltages. I don't trust modern LEDs because their properties tend to vayr with manufacturer and series. This leaves me with only standard diodes, like 1N4148, with 3 or 4 in series Agreed?
Our cable is 1966 and is 4 wires of which only 2 are used. It was dug up when building out from the house. Housed in a hose pipe. It looks so bad as to be surprising it works. tI runs 5 computers all at reasonable speed. Seeing as bandwidth was said to be 2 kHz in the time installed it is doing OK. I watch films throught it.
I just made a thing which would use 1N4148 as limiters. I am using red LED's. As there is some voltage drift any better ideas? I can not give it tons of current and the 1.6 V of an LED as much as I want. I suspect LED's are the best? I think I did notice they react to light. Sad as seeing them work heleps. Black sleeve to cover.
I just made a thing which would use 1N4148 as limiters. I am using red LED's. As there is some voltage drift any better ideas? I can not give it tons of current and the 1.6 V of an LED as much as I want. I suspect LED's are the best? I think I did notice they react to light. Sad as seeing them work heleps. Black sleeve to cover.
It's a pretty large topic actually. Yes standard diodes are reliable and predictable, but if one is after a constant current with temp a string of them will mean the current output will fall with increasing temperature. Also the current will depend to some extent on the diodes' current, although with regulated rails this isn't an issue.
I've seen current sinks for bipolar NPN LTPs use two diodes, and the examples cite the reduction of tail current in the LTP to temperature-compensate the increase in transconductance of the LTP, thus stabilize open-loop gain.
I haven't looked at his site for a while, but I believe Walt Jung was doing a comprehensive survey of current sources, including how high the output impedances. Probably a good place to start.
I was going to use a blue LED as a level shift and our discussion and my measurements of noise made me change my mind and substitute a 3.9V 500mW zener. Also I couldn't find my stash of blue LEDs
It's a fairly noncritical location anyway, but something with fairly low impedance was needed. I'm not too concerned about temperature stability.Likewise here, Brad. When all is said and done, the zener still comes up the winner, in conjunction with a regular diode for temp stabilization, which it actually doesn't NEED, but it can't hurt, and at their currant prices, it's no extra expense to speak of. A 22uF or 47uF cap will clean it all up to make it become a perfect crime.
If the power supply rails are tightly regulated one can also use a simple filtered voltage divider.Likewise here, Brad. When all is said and done, the zener still comes up the winner, in conjunction with a regular diode for temp stabilization, which it actually doesn't NEED, but it can't hurt, and at their currant prices, it's no extra expense to speak of. A 22uF or 47uF cap will clean it all up to make it become a perfect crime.
IIRC what I show in a thread about linuxguru's buffer stage is a composite high-performance current source with a Boxall pair for mostly temperature compensation fed from a voltage divider, and a Aldridge-style driven cascode at the output, for still-higher output impedance. Only problem with it is the need for some high-frequency termination for stability. But noise in the current is mostly from the emitter resistor in the main Boxall device, and this can be reduced further by having more voltage across a larger resistor, the current noise going as the square root of the reciprocal of the resistance.
Hi Nigel,
Reverse biasing signal transistors will degrade Hfe and noise performance for certain. A practical example is an amplifier that goes DC, reverse biasing one transistor of the diff pair. If you see enough of these, you will run into enough evidence as long as you are measuring the beta match on the diff pair. You might even hear the noise, but often it shows as a residual from a distortion test. When you can hear them normally, a lot of damage has been done.
-Chris
Okay, I use red LEDs as references for current sources all the time. They work great, and there was a study done for how noisy they aren't (a few years ago now). Turns out the older red ones are the best. Zener diodes can be more noisy, and as mentioned, the red LED closely matches a pn junction tempco.
Reverse biasing signal transistors will degrade Hfe and noise performance for certain. A practical example is an amplifier that goes DC, reverse biasing one transistor of the diff pair. If you see enough of these, you will run into enough evidence as long as you are measuring the beta match on the diff pair. You might even hear the noise, but often it shows as a residual from a distortion test. When you can hear them normally, a lot of damage has been done.
-Chris
Dan. That even looks like my grass, certainly the wires. If you put OX20 1NS into Google Earth you can see my shed backing on to the football pitch and the stepping stones to the shed. I am 3 rd house looking down the close. That's a bit of fun if anyone can be bothered. All the madnes in the well behaved world comes out of that shed. I think Google have my bottle green Golf Estate. 13 years old now.
For CCS if a slight loss is allowable the true zeners ( above 3.3 V ? ) seem good. I never remember Douglas Self saying CCS could be affected by Early effect. To the best of my understanding that is using the transistor too close to it's turn on point. It gives second harmonic which isn't a big problem. I substuted an LM317 in a cathode circuit of an EL 34 valve. The LM317 with a single resistor forms a current sink. The sub a BD139 and 2 x 1N4148 bias. The result was more second harmonic. With 3 x 1N4148 it was exactly as LM 317. It was running 67 mA in each case. I strongly disapprove of an LM317 in valve circuits and LED's. The LM317 although one of my favourite pieces of electronics is like a TIP3055 and LM741 combined. The BD 139 > 100 MHz. If low frequecncy distortion is identical as is price why use a device 100 times slower. Also once correctly biased the BD139 is mosly a resistor in it's linearity. LED's for small valve cathodes. Why on Earth would you do that ? A resistor is better, as the valve ages more so.
For CCS if a slight loss is allowable the true zeners ( above 3.3 V ? ) seem good. I never remember Douglas Self saying CCS could be affected by Early effect. To the best of my understanding that is using the transistor too close to it's turn on point. It gives second harmonic which isn't a big problem. I substuted an LM317 in a cathode circuit of an EL 34 valve. The LM317 with a single resistor forms a current sink. The sub a BD139 and 2 x 1N4148 bias. The result was more second harmonic. With 3 x 1N4148 it was exactly as LM 317. It was running 67 mA in each case. I strongly disapprove of an LM317 in valve circuits and LED's. The LM317 although one of my favourite pieces of electronics is like a TIP3055 and LM741 combined. The BD 139 > 100 MHz. If low frequecncy distortion is identical as is price why use a device 100 times slower. Also once correctly biased the BD139 is mosly a resistor in it's linearity. LED's for small valve cathodes. Why on Earth would you do that ? A resistor is better, as the valve ages more so.
This circuit costs less than £1. It only needs to be 1% THD if driving a motor. As there is a spare section I have a almost free of charge a lower distortion output. With TL074 I can get > - 70dB. Also it works to calculated values!!!! LED's allow a very healthy output. + 3 x 910 K would give circa 45 RPM in paralell ( Circa 67.5 Hz ). It is bounce free and better than many all in one sine square triangle generators below 100 kHz. Notice it has Sine, Cos, and low distortion outputs.
The Early effect is noticeable in virtually all transistor circuits that have significant voltage swing at the collector. Hence the significant increase in output impedance at the collector of a cascode. However, the collector-base impedance of a bipolar at a given operating point is limited. And even using a JFET or DMOS part as the cascoding device is limited for a simple cascode due to the capacitances. However, if the Aldridge cascode or bootstrapped cascode is used, the effective output capacitance can get very small indeed. Translation here for those used to the conventional attributions: Aldridge-like circuits include the Hawksford cascode, with Malcolm Omar getting audio folk on board many years later.
The only significant disadvantage of the "standard red" LED as the base reference for a bipolar current source is the fairly small available voltage (|Vled - Vbe|) of order 900mV, hence requiring that the resistor in the emitter be fairly small for a given current as well, and thus having higher current noise. The knee of the LED as it were is not so sharp as a medium-voltage zener, but with stable power rails this is not an issue. As mentioned the HLMP-6000 parts are quite dependable and consistent.
But with those reservations, they are very useful, and depending on where such current sources are in the signal chain the noise may be negligible---and you have the ability to swing more volts at the output of the generator for a given power supply rail. A favorite version of the diamond buffer I use has such current sources, as well as bootstrapping the input device collectors to the output emitters. With a smattering of resistors and some high-frequency output devices the resulting six-transistor circuit has exemplary performance, including very low input capacitance (indeed negative at high frequencies, requiring a little lumped C to common, usually supplied by strays and the circuit feeding it).
The only significant disadvantage of the "standard red" LED as the base reference for a bipolar current source is the fairly small available voltage (|Vled - Vbe|) of order 900mV, hence requiring that the resistor in the emitter be fairly small for a given current as well, and thus having higher current noise. The knee of the LED as it were is not so sharp as a medium-voltage zener, but with stable power rails this is not an issue. As mentioned the HLMP-6000 parts are quite dependable and consistent.
But with those reservations, they are very useful, and depending on where such current sources are in the signal chain the noise may be negligible---and you have the ability to swing more volts at the output of the generator for a given power supply rail. A favorite version of the diamond buffer I use has such current sources, as well as bootstrapping the input device collectors to the output emitters. With a smattering of resistors and some high-frequency output devices the resulting six-transistor circuit has exemplary performance, including very low input capacitance (indeed negative at high frequencies, requiring a little lumped C to common, usually supplied by strays and the circuit feeding it).
- Status
- Not open for further replies.