have you tried the linear tech lt3092 for ccs? I've been meaning to try it out . its a 2 terminal low noise low dropout adjustable CCS
The LT3092 is a versatile part for many apps but it doesn't appear to be either high impedance or low noise, compared to more conventional approaches. And although the noise can be reduced by bypassing Rset, they point out that in many applications this will introduce another significant pole and require additional components for compensation. And these will further reduce output impedance.
It does make a low-parts-count and reasonably temp-stable source, but definitely has significant drawbacks.
It does make a low-parts-count and reasonably temp-stable source, but definitely has significant drawbacks.
Hi Patrick, if I may comment, I too have found that putting the i/v resistor AFTER the coupling cap will reduce distortion. This was also shown by Jocko Homo ages ago:
http://www.diyaudio.com/forums/digital-source/6121-easy-build-i-v-stage.html#post61496
The bjt i/v will have lower input impedance.
Best
Alex
http://www.diyaudio.com/forums/digital-source/6121-easy-build-i-v-stage.html#post61496
The bjt i/v will have lower input impedance.
Best
Alex
oh yeah sure, its not sub nV low noise, so your right its not ideal, but i do think its a part worth exploring for audio that hasn't really had any airtime here, which surprises me given how versatile it is. I did note the outputZ issue, thanks for the feedback, i still think i'm gonna grab a couple to play with.
In the LT3092, 4 factors determines the output current thermal stability :
1. the reference current source
2. the current setting resistor
3. the output resistor
4. the opamp input offset voltage
Taking the values from the performance graphs on P.4 of the data sheet, the thermal coefficients of (1) is about 20ppm/K.
That of (4) is about 4ppm/K, assuming 2V voltage drop across the resistors.
We can assume that we use good resistors of say 5ppm/K for both 2 & 3.
The worst case thermal drift for (1)~(4) combined is ~ 30ppm/K.
On the other hand, the thermal drift of the MAX6126 current source depends on the tempco of the voltage reference (3ppm/K) and that of Rout (5ppm/K).
It is not difficult to see which has better thermal stability.
As to output current noise, the MAX6126 has 1.3µVp-p for 2.048V. Assuming we need a bias current of 2mA, then the current noise (excluding Rout noise) is 1.27nAp-p. If my sums are correct, the LT3092 has about 200nA noise for 2mA output current.
And the MAX6126 requires one reference resistor less.
So it still gets my vote.
Patrick
.
1. the reference current source
2. the current setting resistor
3. the output resistor
4. the opamp input offset voltage
Taking the values from the performance graphs on P.4 of the data sheet, the thermal coefficients of (1) is about 20ppm/K.
That of (4) is about 4ppm/K, assuming 2V voltage drop across the resistors.
We can assume that we use good resistors of say 5ppm/K for both 2 & 3.
The worst case thermal drift for (1)~(4) combined is ~ 30ppm/K.
On the other hand, the thermal drift of the MAX6126 current source depends on the tempco of the voltage reference (3ppm/K) and that of Rout (5ppm/K).
It is not difficult to see which has better thermal stability.
As to output current noise, the MAX6126 has 1.3µVp-p for 2.048V. Assuming we need a bias current of 2mA, then the current noise (excluding Rout noise) is 1.27nAp-p. If my sums are correct, the LT3092 has about 200nA noise for 2mA output current.
And the MAX6126 requires one reference resistor less.
So it still gets my vote.
Patrick
.
Last edited:
haha yeah just a little different, but like i said just a bit of fun and its the versatility and simplicity that i like; plus i'm just addicted to linear tech, coz i love the fantastic app notes.
i've got some of the LTC6655 on the way too 0.25ppm p-p 0.1hz-10hz noise, 2ppm/c max drift. that plus lme49990 and lme49610 will be the new clock reg for my portable. been doing lots of battery power stuff, thus the LT love as efficiency is off the charts generally
i've got some of the LTC6655 on the way too 0.25ppm p-p 0.1hz-10hz noise, 2ppm/c max drift. that plus lme49990 and lme49610 will be the new clock reg for my portable. been doing lots of battery power stuff, thus the LT love as efficiency is off the charts generally
You would think a resistor is needed there, but really it isn´t. Just a little tweaking of the current sources and it works and sounds fine.
Best
Alex
So this should work, but you have to try, as I don't have a TDA1541 handy.
The bias current of the MAX6126 itself is about 0.5mA, but it is temperature stable (about 1uA/°C). It is also stable enough against Vin. So in theory, you can wire the current source circuit such that the collector of the BJT and Vin of the MAX6126 is connected to the +ve power supply (say +5V). Then connect the "Gnd" of this circuit to the "Load" -- the current sink of TDA1541. Since the output pin of the TDA1541 should be at, or close to 0V, this will provide a virtual Gnd for the MAX6126 current source. The current through the BJT has to be reduced by the bias current of the MAX6126.
Else we can also make up a similar circuit using a -ve rail voltage reference (like TL431). But they are by no means as high spec.
Or try Jeremy's solution with the LT3092.
Patrick
The bias current of the MAX6126 itself is about 0.5mA, but it is temperature stable (about 1uA/°C). It is also stable enough against Vin. So in theory, you can wire the current source circuit such that the collector of the BJT and Vin of the MAX6126 is connected to the +ve power supply (say +5V). Then connect the "Gnd" of this circuit to the "Load" -- the current sink of TDA1541. Since the output pin of the TDA1541 should be at, or close to 0V, this will provide a virtual Gnd for the MAX6126 current source. The current through the BJT has to be reduced by the bias current of the MAX6126.
Else we can also make up a similar circuit using a -ve rail voltage reference (like TL431). But they are by no means as high spec.
Or try Jeremy's solution with the LT3092.
Patrick
Perhaps a Wilson mirror using PNPs and with sufficient emitter ballasting to ensure low noise, fed from the 6126-based current sink?
Dear All,
I'm trying to match the BJT using this circuit, at the end of this page
Transistor matching
Now, on Vc I manage to be accurate, but Vb is around 2mV and in this range my multimeter has only one digit. So what I obtain is that the but artificially group around Vb.
Can I increase Rb by a factor of 10 ? Or maybe is better just to group them by Vc.
Thanks,
Davide
I'm trying to match the BJT using this circuit, at the end of this page
Transistor matching
Now, on Vc I manage to be accurate, but Vb is around 2mV and in this range my multimeter has only one digit. So what I obtain is that the but artificially group around Vb.
Can I increase Rb by a factor of 10 ? Or maybe is better just to group them by Vc.
Thanks,
Davide
Davide,
Many DMMs have hfe measurements built in.
You just need to plug in a BJT and will get a value.
Since you are using a high hfe part, it will give you 3 digits (e.g. 453).
So your uncertainty is well below 1%.
Failing that, you can also match using the circuit in P.7 of the data sheet of BCM846s.
It needs a bit of setting up, but is not so difficult.
For the current source, you have to calculate the required current by :
I_ccs = 2x I_bias / hfe
So if your BJT should each see 4mA bias, and you expect hfe of say 500, then I_ccs = 16µA.
Hope it is clear,
Patrick
Many DMMs have hfe measurements built in.
You just need to plug in a BJT and will get a value.
Since you are using a high hfe part, it will give you 3 digits (e.g. 453).
So your uncertainty is well below 1%.
Failing that, you can also match using the circuit in P.7 of the data sheet of BCM846s.
It needs a bit of setting up, but is not so difficult.
For the current source, you have to calculate the required current by :
I_ccs = 2x I_bias / hfe
So if your BJT should each see 4mA bias, and you expect hfe of say 500, then I_ccs = 16µA.
Hope it is clear,
Patrick