I'm getting more and more into low noise design nowadays since being surprised by how much difference a lower noise I/V stage made on my DAC's sound a year or so ago. Prior to that discovery I just assumed that if the (calculated or sim'd) noise was a few dB below the recording noise floor I'd be fine. Wrong!
In order to have more of a handle on noise I'm wanting to venture into making some noise measurements, something I've not done before. For which I reckoned I needed to acquire a couple of things - a low noise preamp and a lowpass filter (because absent a defined bandwidth noise measurements aren't worth much). On the preamp, I bought a 100X gain AD8429 instrumentation amp PCB from Taobao - this needed a PSU so I built one. But then when I realized I'd also need a lowpass filter I felt in my habitual laziness this was becoming a bit too much of a work-up and I felt a passive solution beckoning me. A step-up transformer.
A trafo has some nifty advantages. First its capable of lower noise than any opamp known to man seeing as its limitation is the resistance of the windings which can easily be below 50ohm which is about the equivalent limit of the best opamps (LT1028, AD797). Second it can provide the necessary low pass filtering if its leakage inductance is correctly sized. The primary disadvantage is sensitivity to external fields so it might end up needing its own screened box.
The highest ratio transformer in my handy spares box is 80:1, this was an EE35 step-down designed for a DAC output stage that's still awaiting a functioning DAC front end. So I decided to re-purpose it into a step up. Its not ideal but it has the considerable advantage of being ready to run. The first thing I wanted to understand was how to get a reasonably flat frequency response out of it. In step-down operation this has never been an issue but step-up is a little bit trickier. Fiddling the trafo's output loading by trial and error is a thankless task so I turned to LTSpice to get me in the ballpark. For which I needed some measurements - inductances and resistance. With those values taken and plugged into the model then I was able to come up with a 'zobel network' on the secondary giving a reasonably flat response, but only up to 4kHz or so after which it rolled off.
The roll-off is due to the fact that this trafo was designed to handle a rather high primary voltage and hence has a huge (for a ferrite) primary inductance of over 1000H. That much wire (its total resistance is about 4.5kohm) brings with it a lot of parasitic capacitance which does strongly limit the bandwidth in step-up mode. To get to the full audio bandwidth I reckon I need a leakage inductance 4X lower. I do also wonder if this trafo might be much bigger physically than necessary - seeing as its not going to have any signal to speak of passing through it, it barely needs any volt-seconds. I might still have some EP17s from my 'GrossDAC' project around somewhere if I could be bothered to institute a search. For now though I got some noise readings from this EE35 which make some kind of sense - at least they did after I put it in a lunch canteen to partially screen it.
A single LM317 returned about 19uVRMS at the lowest possible output voltage (1.25V). [Bear in mind the bandwidth is about 5kHz and the FR has a boost prior to the roll-off around 3dB). Putting zener diodes on to pin1 to increase the output voltage didn't increase that noise at all when they were bypassed with 560uF capacitance. This shows that zeners (here I used 2 * 10V in series to get 21.25V) don't significantly increase the noise over the internal bandgap. A satisfying result but I'm hankering after something a little lower noise than a bog-standard LM317.
A few weeks ago I prototyped a discrete shunt using a Sziklai pair of transistors and got wifey to turn it into a PCB. The noise reading from this was a huge disappointment at almost 80uVRMS. After a minute of headscratching I wondered if I was running the zeners too current-starved. So I jacked up their bias to 5mA from the original 1mA. At 5mA the noise reading was 15uV, a worthwhile improvement over the LM317. Higher currents reduce the noise further still. Schematic below - note the 1k resistor values are place-holders.
In order to have more of a handle on noise I'm wanting to venture into making some noise measurements, something I've not done before. For which I reckoned I needed to acquire a couple of things - a low noise preamp and a lowpass filter (because absent a defined bandwidth noise measurements aren't worth much). On the preamp, I bought a 100X gain AD8429 instrumentation amp PCB from Taobao - this needed a PSU so I built one. But then when I realized I'd also need a lowpass filter I felt in my habitual laziness this was becoming a bit too much of a work-up and I felt a passive solution beckoning me. A step-up transformer.
A trafo has some nifty advantages. First its capable of lower noise than any opamp known to man seeing as its limitation is the resistance of the windings which can easily be below 50ohm which is about the equivalent limit of the best opamps (LT1028, AD797). Second it can provide the necessary low pass filtering if its leakage inductance is correctly sized. The primary disadvantage is sensitivity to external fields so it might end up needing its own screened box.
The highest ratio transformer in my handy spares box is 80:1, this was an EE35 step-down designed for a DAC output stage that's still awaiting a functioning DAC front end. So I decided to re-purpose it into a step up. Its not ideal but it has the considerable advantage of being ready to run. The first thing I wanted to understand was how to get a reasonably flat frequency response out of it. In step-down operation this has never been an issue but step-up is a little bit trickier. Fiddling the trafo's output loading by trial and error is a thankless task so I turned to LTSpice to get me in the ballpark. For which I needed some measurements - inductances and resistance. With those values taken and plugged into the model then I was able to come up with a 'zobel network' on the secondary giving a reasonably flat response, but only up to 4kHz or so after which it rolled off.
The roll-off is due to the fact that this trafo was designed to handle a rather high primary voltage and hence has a huge (for a ferrite) primary inductance of over 1000H. That much wire (its total resistance is about 4.5kohm) brings with it a lot of parasitic capacitance which does strongly limit the bandwidth in step-up mode. To get to the full audio bandwidth I reckon I need a leakage inductance 4X lower. I do also wonder if this trafo might be much bigger physically than necessary - seeing as its not going to have any signal to speak of passing through it, it barely needs any volt-seconds. I might still have some EP17s from my 'GrossDAC' project around somewhere if I could be bothered to institute a search. For now though I got some noise readings from this EE35 which make some kind of sense - at least they did after I put it in a lunch canteen to partially screen it.
A single LM317 returned about 19uVRMS at the lowest possible output voltage (1.25V). [Bear in mind the bandwidth is about 5kHz and the FR has a boost prior to the roll-off around 3dB). Putting zener diodes on to pin1 to increase the output voltage didn't increase that noise at all when they were bypassed with 560uF capacitance. This shows that zeners (here I used 2 * 10V in series to get 21.25V) don't significantly increase the noise over the internal bandgap. A satisfying result but I'm hankering after something a little lower noise than a bog-standard LM317.
A few weeks ago I prototyped a discrete shunt using a Sziklai pair of transistors and got wifey to turn it into a PCB. The noise reading from this was a huge disappointment at almost 80uVRMS. After a minute of headscratching I wondered if I was running the zeners too current-starved. So I jacked up their bias to 5mA from the original 1mA. At 5mA the noise reading was 15uV, a worthwhile improvement over the LM317. Higher currents reduce the noise further still. Schematic below - note the 1k resistor values are place-holders.
