Finding voltage regulator for high speed analog circuit

[dzseki]

Given a complex system with various circuits and functions. The whole system is running off a common power supply with having +/-15V and +/-5V etc. A part of this system is supposed to be a high performance, high speed (~250MHz) analog amplifier section. Where low distortion and „clean” signal amplification are the key features.
At the moment this analog section is using the common +/-5V rail, but opamps of this circuit are decoupled with R-C filters like 4R7-10uF or 1R-100nF, but no coils or ferrites throughout the power rails anywhere.
Do you think would it worth to give this circuit a separate local +/-5V to decrease cross coupling with other circuits and so on?
If so, what would be the main point of consideration in choosing the regulator. I mean, what would modern regulators offer over a conventional 7805 for example in this application?

[iMo]

The decoupling of high speed precision circuits is difficult.
The capacitor's impedance is V shaped with rather narrow range where their impedance is "low" (ie <1ohm). And you need quality smd ceramics.

To cover a broader decoupling range (in your case say up to 1-2GHz) you would need several capacitors, with different values, wired in parallel.

See for example these posts on the topic:
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg2375868/#msg2375868

Moreover, for your app the pcb layout (shapes/lengths of the pcb tracks, number of vias, ground loops, etc) plays a significant role (because the parasitic inductancies make decoupling harder and may cause instabilities).

Therefore the design of such an 250MHz amplifier is similar to a design of an UHF circuit.

The classic 7805/15 and 7905/15 have usually much larger low freq noise (in xHz-100kHz range) against some modern Vregulators.

[dzseki]

Thank you.

This is not my design, and I have no intention to rebuild/redesign as a whole, I was just asked to improve the circuit further if I can... I guess I have to live with whatever PCB layout it has.

Although adding more capacitors seems to be a good idea. In fact when I did circuit design in this field I usualy used a combination of four SMD capacitors for power supply bypassing in the order of 10nF, 100nF, 1uF, 10uF.

Using a filtered zener diode with emitter follower as a local power supply would help in pushing down the LF noise, to my understanding bandgap references (those are found in most of the 3 pin regulators) are noisier than zener diodes. Voltage stability over time/temperature is of less concern here.

[David Hess]

At the moment this analog section is using the common +/-5V rail, but opamps of this circuit are decoupled with R-C filters like 4R7-10uF or 1R-100nF, but no coils or ferrites throughout the power rails anywhere.

...

Do you think would it worth to give this circuit a separate local +/-5V to decrease cross coupling with other circuits and so on?

The only reason I would do that is if I required better regulation at low frequencies or lower low frequency noise.  At higher frequencies, the RC (or RLC) decoupling is better than any individual regulator anyway; if you need better decoupling or better load regulation or both, then I would first add inductors in parallel with the resistance of your RC decoupling networks.

Some circuits like low phase oscillators and low jitter logic benefit from precision local regulation which takes more of the form of a power reference.

If so, what would be the main point of consideration in choosing the regulator. I mean, what would modern regulators offer over a conventional 7805 for example in this application?

Within limits for a given transient response, the regulator's speed can be traded off for the amount of output bulk decoupling.  So faster regulators require less output capacitance for a given transient response and this can become important in high dI/dT applications like high performance CPUs where it may be difficult to get enough capacitance in place with a low enough impedance.

Low frequency noise is difficult to filter out so it also benefits for a higher performance regulator as mentioned above.

[T3sl4co1l]

Impossible to say -- what is the PSRR of the various stages, at whatever frequencies are relevant?

Tim

[dzseki]

Impossible to say -- what is the PSRR of the various stages, at whatever frequencies are relevant?

Tim

The circuit is built around current feedback opamps (CLC449) and analog multipliers (AD835). The CLC449 has a PSRR of >50dB to 1MHz. The AD835 has 0.5%/V PSRR (~50dB) at lower frequencies, but has a peak around 5MHz of 10/20dB (V+/V-), then it gets better again.