Some improvements to the basic CCS's have been described here:
http://www.diyaudio.com/forums/solid-state/173005-improved-current-source-sink.html
Here are some further circuits, a bit more complicated but having a very wide compliance.
On the left is the basic circuit.
The performances are already quite good, with a variation of 0.25mA (for 10mA nominal) for a voltage swing of 2.5V to 62.5V.
The middle circuit has two improvements: the 22Meg, to avoid relying on leakage currents for start-up, and a 300K compensating resistor decreasing the delta current to ~25µA, 1/10th of the initial value.
If the tempco of -0.3%/°C is problematic, each pair of silicon diodes can be replaced by an LED, providing a first order compensation.
The circuit on the right uses the infrared LEDs of an optocoupler to minimize the working voltage.
http://www.diyaudio.com/forums/solid-state/173005-improved-current-source-sink.html
Here are some further circuits, a bit more complicated but having a very wide compliance.
On the left is the basic circuit.
The performances are already quite good, with a variation of 0.25mA (for 10mA nominal) for a voltage swing of 2.5V to 62.5V.
The middle circuit has two improvements: the 22Meg, to avoid relying on leakage currents for start-up, and a 300K compensating resistor decreasing the delta current to ~25µA, 1/10th of the initial value.
If the tempco of -0.3%/°C is problematic, each pair of silicon diodes can be replaced by an LED, providing a first order compensation.
The circuit on the right uses the infrared LEDs of an optocoupler to minimize the working voltage.
Attachments
Last edited:
That is an impressive result, but of course, this is not what these threads are about.
I have described a number of real-world solutions to implement 2W. current sources (that is, fully floating).
It is obviously possible to do much better if you drop the 2W. requirement, and to do even better if one is allowed to use "sim components", like perfect floating sources.
If that route is allowed, there is an obvious shortcut, giving even better results, see below: voltage range=anything, Inom=10mA, deviation=0.
The circuits I propose are shown in a simulated form, but they can readily be built using normal, physical components, and I am confident the discrepancies between the simulated and actual versions will not exceed several percents.
Simulators are primarily intended as a means to an end, but they can also be used for phantasy and intellectual masturbation.
This is not what I am doing here.
That said, the circuit you propose is not without merit, and for that matter can be found in textbooks.
But it would need some adaptations to qualify.
First, the way it is drawn is misleading: in fact the 5V source is in series with the stimulus, which means that in fact the actual voltage swing is from 7.5 to 67.5V.
The elements could be rearranged in a more normal manner, to avoid having the main current passing through the 5V bias.
This would limit the minimum operating voltage to ~5V, and then would remain the problem of creating the 5V bias from the voltage available, without disturbing the controlled current.
Certainly achievable, but not simple or straightforward.
But anyway, I rule out the use of FETs. They may look great in sims, but they suffer from big dispersions in reality.
The Idss of the 2N3819 can range from 2 to 20mA. This means that you have to discard first hand half of your samples, and then adjust individually each circuit for the rest.
Even if you resort to sorted devices, like the BF245A, B or C, there is still a dispersion range of more than 1 to 2 in the Idss.
BJT's may give somewhat lower performances, but the current and tempco are predictable, and in most instances they won't require any adjustment or sorting.
Attachments
Last edited:
Yes, of course, and the tempco will also be cancelled to the first order.
Ideally, it should be tuned to the Early effect of the transistors you use and the emitter resistors, but you can dispense with it completely: that's what is done in the first example, but of course, you can notice that the current variation for a 2.5 to 62.5V voltage range is much larger than for the two next ones: ΔI(R7)=250µA, but ΔI(R8 or 13)~=30µA, that's one order of magnitude improvement
Hi Elvee
I am studying your solution that I find really good.
How do you calculate the value of R3 (300k) ?
What is the delta current ?
Best
Ricardo
R3 compensates for Early effect of the transistors. The Early voltage is never published in the datasheets though, and the value found in the spice model is generally very simplistic and crude: for example, many low-power ONsemi transistors have the same 100V value.
It has to be determined experimentally if you want an optimum performance, and for ultimate performance it would need to be trimmed according to the components used.
The delta current is the difference between the lowest and highest test voltages: the lower, the better
This circuit is just a building-block, to be used as a component of a complete circuit. To make it widely variable would be complicated as it requires the simultaneous variation of 2 resistors, maybe three if you want good performance.
What you need is something like this:
https://www.diyaudio.com/community/threads/dc-bias-feed-for-inductors.311944/post-5177420
Of course, this one has all the bells and whistles you can imagine, but you get the idea: you could prune it to tailor it to your needs
I like the original improved CCS best, no PNP, no JFET, ... JFETs are cool except they are unpredictable. Perhaps we need to use them with a POT in the source, but POTs and resistor selection is not acceptable for commercial production. The biggest problem with the RJM1 circuit is the 5V supply. A two-wire solution is important because it does not require anything of the rest of the circuit. The CCS etc is not an end unto itself. How many chips are useless because they require more support circuitry than they replace.