What's the PSRR of an emitter follower?

I asked this question of Yahoo and didn't find anything much - a few people mentioning the PSRR of their C-multipliers but no simulation results and no algebra. So I fired up LTspice for myself to take a look. The transistor models are the usual Gummel-Poon ones LTspice provides (one up from hybrid-pi) which look to be decent enough for this purpose.

I went for two EFs (sometimes called EF2) as that's probably a more practical arrangement in an audio amp. Some designers even prefer EF3 to get much lighter loading on the VAS/TIS. The EF load I made independent of the bias current so I could learn more about biassing. I used AC simulation to have a look at how the PSRR varied over the audio band.

Turns out the PSRR depends on at least three circuit details and one inherent characteristic of the EF transistor used. In no particular order the circuit aspects which matter are the source impedance seen by the base of the EF, secondly the load impedance seen and finally the bias (emitter) current.

The transistor figure of merit which trumps most of the circuit details is the Early voltage - in Spice its termed 'VAF' meaning 'forward Early voltage'. Its a measure of the collector output impedance - if the curves of iC vs VCE at varying iB show horizontal lines you've got a high VAF transistor and can expect decent PSRR from your EF. Sloping lines are bad news for PSRR.

I like intuitive ways of getting a grip on what's going on - how I do this for the EF is it has an input impedance (hoe looking into its collector) and an output impedance (1/gm looking into the emitter). To a first order the PSRR is the ratio of these two but both decrease with increasing iC (hence iE).

Overall though it was rather a shock to me that the PSRR could be so bad. But then again it can look quite good given judicious choice of EF transistor and its model. For example Bob Cordell's model for the MJE15032 has VAF=2000 which is an amazingly high figure. He must have measured real transistors as the DS doesn't give any indication how this figure could be derived - I need to ask him directly how he got it so high. Perhaps there's an error somewhere because I was under the impression that the king of VAF was the 2SC3503. Bob's model for this has it has 'only' 769 which seems very reasonable as the iC lines look amazingly flat in the DS all the way up to 100V.

The outcome of my short time running sims has made it obvious to me that the simple PSRR plot shown for my LM4766 cannot reflect the real world application - simply because no load resistance is specified and no output voltage (hence EF emitter current) is specified. Both these make a difference to the PSRR in practice. We can fairly safely assume the PSRR in the DS was simmed with no signal applied and with infinite load impedance attached.

Since composing this but before publishing I found this reference on Bob Cordell's power supplies discussion thread : https://www.diyaudio.com/forums/solid...ml#post2984622
where a poster did a bare-bones analysis for an EF. jcx disagreed with his result. The mention of EF OPS PSRR didn't generate any discussion.

I realized I forgot to mention one thing about how PSRR varies that came to light. It gets considerably worse as the EF nears saturation - i.e. the output gets closer to the supply rail. There are two effects - the hoe is lowering so PSRR independent of freq goes down. But also the transistor's Ccb gets greater so the PSRR degrades more at higher frequency at higher outputs. This makes sense of subjective observations of SQ - I notice on piano transients (for example) that the die-away is affected by rail capacitance. It also fits with Frank's subjective description of the sound 'falling apart' at higher levels - even if an amp isn't actually clipping, if its spending a larger proportion of its time near the rails the sound's going to get muddier, particularly at HF if the dominant factor affecting SQ is the EF's PSRR.

Attached Thumbnails