Power Supply and Linear Regulator Design for RF
To introduce myself, I’m the primary founder of Holzworth Instrumentation. I have the usual years of background at other companies, too many degrees and generally overly educated for my own good. What all this means is that I’ve learned some very valuable lessons – some of the most useful things related to circuits, I’d learned in my Junior year while completing my BSEE. These circuit design techniques and linear regulator design are the most useful skills (I believe) an RF engineer can have.
At my first job out of school, all of my focus was on microwave design – assuming that it had to be the good stuff. Time and time again, the RF only took half the time while the other half was designing the low noise circuitry with Op-amps and BJTs; which are really what make the RF work after all. I was fortunate to have the remaining professor at Oregon State without a PhD, teach me data acquisition. He did not teach from text books – he was the text book. I remember him laughing at us during our Senior year because none of us could really use a BJT very well. We laughed too, mainly because he was right. I learned more useful design skills from that class than most of my post-graduate classes.
Now fast forward more than a few gray hairs, we were designing our very first synthesizer – the HS1001A. Care to guess what circuit that took the most design time? Power supply. Not the AC-DC kind, but the standard old linear regulators. What I found was that even the ‘low noise’ linear regulator ICs weren’t very low noise. By this time in my career I’d had an intensive phase noise background where everything was running off batteries to maintain extremely low noise levels. But how to make a linear regulator that is close to the noise of a battery? Was it possible to take whatever a customer applied as a source voltage and make all the noise disappear to the level of a battery?
Enter the noise stuff. Everyone remembers Boltzmann’s constant, or at least hearing about it. And then this ’Johnson noise’ stuff, and somewhere during that Junior year you learned that resistors make Johnson noise (white noise) depending on the resistance value of the resistor in the form of noise = sqrt(kTBR). I’ll keep equations to a minimum as Steven Hawking correctly pointed out that the number of readers is inversely proportional to the number of equations. That notedequation also relates that the bandwidth you look at is as important as the temperature. The temperature is in Kelvin and for practical
purposes, fairly constant near room temp. The phase noise guys really prefer everything normalized to a 1Hz bandwidth, so that the ‘B’ variable goes away. The equivalent resistance (‘R’) is really what
causes noise in our system.
What level of noise? A 50ohm resistor (or system) has around 0.9nV/rt(Hz) of noise. A 1k resistor has ~4nV/rt(Hz) of noise and a 100k resistor has ~40nV/rt(Hz) of noise. When you look in terms of a log
scale (using 20log because of voltage) you can see a 1k resistor has 12dB more noise than a 50ohm resistor and 100k has 20dB more than the 1k. These are huge in terms of microwave. I bring this up because these are the numbers you’ll see on those linear regulator and op-amp spec sheets you might use to build your own linear regulator.
You’re designing RF now – why would all this matter? Well, let’s take an amplifier as an example example, an LO amplifier that is pretty critical to system performance, maybe you’re driving a mixer with it. You want to drive it into p1dB compression to suppress AM noise, a clever trick thinking you’re getting rid of AM noise, but are you? Ask now, what is the AM-PM conversion of this amp? Most engineers have never even measured it. It’s pretty simple. Take the amp and put it on a network analyzer, measure the S21 phase at your nominal power supply, now drop or raise the power supply a volt (or something) and measure it again – it’s different right? This is AM-PM conversion. Any noise on your supply will be converted into phase noise in a nice neat step. And once it’s there, you can’t get rid of it.
Now say you hand off the RF amp’s linear regulator design to the ‘new engineer’, a process I find baffling. Nothing against the new engineer, but we typically don’t learn good regulator design in school. He/she uses an average linear regulator IC that has 100nV/rt(Hz) of noise on it (40dB above 50ohm noise), job done. When compressing an amplifier, the voltage swings almost entirely across the I-V curve, causing fairly large changes in phase. This is why time domain guys use differential amps – they don’t suffer from this, but they have other signal to noise problems that make them not suitable for phase noise work. Say the phase moves 20degrees over a volt change, a ‘volts per radian’ of 0.35 exists, which is the inverse of a mixer as a phase detector. Now multiply by the voltage noise of 100nV/rt(Hz) and take the 20log of it. You get a phase deviation of -150dBc/rt(Hz) with 100nV of supply noise. This is what
you’ll measure for ‘phase noise’ at the output of your amplifier that is probably capable of much better. Worse, your linear regulator is probably much worse than this at 1kHz offsets compared to 100kHz offsets.
So now you’ve just spent all this time on choosing the best mixer, amplifier, oscillator and you degraded the entire system with a $0.50 linear regulator. How good does the regulator need to be? Well that
depends on your desired system performance level, as described above. What does Holzworth strive for? The linear regulators that we design discreetly exhibit between 1nV/rt(Hz) and 3nV/rt(Hz) depending on the application and usually > 100dB rejection from input to regulated output.
How do you go about finding out about linear regulator design? Take a step back, do some web searching on low noise regulators. The audio guys are the ones that really do a good job. Start there. Next to phase noise work, audio has the most use for a low noise regulator.
In the next blog, I’ll discuss how to measure the voltage noise.