For my project “A Noise Figure Measurement Solution for the Poor” using a RIGOL DSA815TG budget spectrum analyzer I needed an external preamplifer for testing. The INA02186 I made came in very handy, except for one little issue:
- how would you power this amplifier so no external line power is necessary that would make it big and bulky ?
- how can you make sure that no RFI and noise coming from the power side are spoiling amplifier performance ?
Click here to see a Noise Figure Measurement Solution for the Poor …
Click here to see an INA02186 MMIC Amplifier …
If you look at the DSA815TG, there is no output that can be used for powering up probes or other devices like on a Keysight N9000A CXA or a E5071C VNA with their “probe power” outlets on the front panel (they offer +15V and -12V at sufficient current rating for what we need). The only thing that could be used is a USB output on the DSA815 front panel (we may assume that a maximum current of ca. 0.5A is OK).
On the other side, the load requirements of the INA02186 preamp are very modest; The input voltage should be ca. 15V at max. 50mA, but the load is very constant and does not have any spikes or pulse currents due to the very low level input signal level of this preamp (-30dBm max.). We may say, the load is more or less a constant resistor paralleled with a capacitor.
But USB ? Really ? USB is a high-speed digital bus renowned for everything except plain and clean power. Still, even the Keysight U7227A 10MHz to 4GHz preamplifier uses it as a power source, so it must be possible to convert this into a clean analog power of ca. 15V. The only question is how much effort needs to be spent for killing off all RF hassle on the analog output side.
USB has 5V, we need ca. 15V@50mA maximum. That means that we need a voltage converter, a rectifier and filter on the output side, a regulator with feedback and some anti-RFI circuitry.
The first try was a tiny, compact, commercial potted 5V/15V converter module. It worked, but the output was infested with a few 10mV of ripple at a few 100kHz and substantial RF garbage from shortwave to beyond the VHF bands. I tried to reduce this down to a manageable amount by lowpass filtering, but I did not find a filter with sufficient stopband attenuation so I could live with the remnants. Either stopband was OK, then filter corner frequency was too high, or corner frequency was OK, but the stopband attenuation deteriorated in a frequency range where spikes and noise residue would affect my preamp. The obvious idea to cascade the filters did not work out perfectly, either, and in the end the whole thing was getting large, too. So, I was out of luck.
After some unsuccesful fiddling, I got the idea to do everything the other way than industrial converter modules want to do it. Their design idea is to
- make switching frequency high so very small ferrite cores can be used
- Efficiency is king, because these modules have no cooler and must generate very little heat so they do not burn out and die.
- use superfast switching transistors to keep losses low
- regulate by PWM so no inefficient linear regulator is needed
- meet RFI requirements by dithering and spectral spreading instead of by “real” RFI reduction
So what I needed for an RFI-optimized converter is something like this:
- make switching frequency as low as practical so keep harmonics low
- Efficiency is uninteresting as long as USB can supply the DC power.
- Switch slowly, use low frequency transistors and snubbers to keep harmonics low even if some losses are the price for this.
- Regulate linearly and continuously
- Kill RF by slow risetimes, low frequency, filtering, a linear regulator and linear postregulation
My design approach was to use a regulated Royer converter with an ISDN transformer, a voltage doubler at the secondary side, followed by an RC filter and a capacitance multiplier. The feedback loop is closed by an optocoupled Zener feedback of the output voltage that throttles the working voltage of the Royer converter. There is also an input current limiter to ca. 0.5A and a soft-start circuit. The principal schematics (not 1:1, the real unit is more complicated) looks like this:
There are some things here that I cannot model accurately:
- The ISDN transformer has no model available, and especially core saturation is not simulated. For a realistic approach, this is a must. Here, external drive replaces the Royer transistor base feedback.
- The regulator feedbeck is tapped off the secondary side using an optocoupler instead of a direct connection.
- Primary and secondary sides of the regulator are galvanically isolated to prevent ground loops.
Please note the use of special ZTX851 transistors in the Royer converter to handle the ample current at low collector voltages and small base drive. These remarkable TO92 transistors can switch up to 5A (!) and have a very high gain even at low voltages and high current.
At the output part, there is no regulator, but a capacitance multiplier. This cleans up ripple a lot better, but it does have a few 100mV drop from no load to full load. This supply is intended for ripple and spike sensitive, but constant loads like MMIC amps or similar devices. Moderate voltage drops under load are unimportant there. Coarse regulation happens via the PC817 optocoupler feedback.
Click here to see the Datasheet of the ZTX851 transistor …
The PCB of my prototype can be seen here:
You can see the different grounding areas. The input part can be either floating, connected to the USB cable shield, or the output and case ground. I had the best results with the USB cable shield grounded to case and the internal conductors floating.
The connectors right behind the USB socket allow the implementation of a “daughter board” where USB could be used to power some processor in parallel to the converter part. Here, everything is jumpered for a direct USB connection, shorting USB D+ to D- indicating that this device will be recognized by the USB subsystem as a USB battery charger device without other functionalities.
The PCB was housed in a tinplate Schubert box with the USB connector on one side an a CINCH connector at the other.
So far, it worked quite well as an isolated supply. I combined it with a low noise preamp and it worked fine, too !
Click here to see a USB Powered Low Noise Preamplifier using this Converter …
Now lets see how the power spectrum at the 15V side looks under full load (300Ohms, translates to 50mA of output current), first from 9kHz to 100MHz:
Up from just a few MHz (we look into this later) everything is below 100dBm. Good enough for sure ! Just to be on the safe side and to make sure that our oscillator is really oscillating, lets go to the very low end (9kHz to 500kHz):
OK, it does oscillate. Oscillation frequency is ca. 4.1kHz. It is a balanced design, so the even harmonics cancel out almost completely (-105dBm). Even below 500kHz, we stay below ca. 90dBm, coming down to less than -100dBm at 400kHz. I could not ask for more, job done.
Just an aside regarding measurement techniques:
Everything, including load resistors, must be coaxial and shielded. Otherwise you will measure everything in your surroundings (radio stations, power line communications, hairdryers, LED lighting, triac regulators, …) and you will not even see such weak signals.
Scope Measurements
On a scope, the 15V@50mA waveform looks like this :
Frequency is 4.16kHz, with 2.135mVpp, 760uVrms spikes on it. When we calculate back, 2.135mVpp translate to ca. -80dBm for the fundamental. In the MMICs we want to power with this one we have another input cap combo plus a voltage regulator, so we can expect less than -110dBm at the MMIC. That should suffice.
Just to be on the safe side, I added a coaxial low pass (3.3Ohms, 3300uF) at the output. This would cause a voltage drop of ca. 165mV at full load, but it has cutoff frequency of ca. 90Hz. After that, the oscilloscope showed a flatline with no visible spikes (no photo). Expected signal level would be about 50uV, into 50Ohms that would be less than -110dBm.
Missing: pulse testing, input current
Electronic Projects for Fun 