I am trying to select a low noise NPN TO-92 transistor for use in a reverb recovery circuit. Source impedance is 2.575K Ω, source resistance is 215Ω, signal level is ~4mV. Using the Horowitz-Hill table for low-noise BJT transistors suggests that a 2N3904 transistor would be excellent for that application, with a voltage noise of 1.9nV√Hz and current noise of 0.68 pA √Hz. But the datasheet for the ON Semi 2N3904 quotes a Noise Figure of 5 db @ 1KΩ, which I calculate translates to 5.9nV√Hz. That is significantly higher than the H-H figures, which would indicate a total noise of 2.02nV√Hz @ 1K Ω. Which one is correct?
Yes, I believe some are, but the old-school ones I'm using are spring reverbs (Accutronics, Belton, MOD). They have that classic sound that I find pleasing.
I take it you have never played a guitar through an amp with a noisy reverb. It's really quite annoying; so YES, it DOES matter.
Although it is FET and thus very low current noise, a TL072 is one of the noisiest opamps available, with voltage noise of 18nV√Hz. That would make it at least 12-15 db noisier than a bipolar transistor in this application.
Last edited:
Piezo reverbs grace some old Silvertones and Danelectros, but they were REALLY poorly implemented and had all the tone of a screen door spring. No one has used a piezo for new construction in decades. Spring reverbs have been the thing ever since Hammond invented tham.
I have played through many noisy reverbs, and offhand I can't think of many that had noisy designs. I often found ICs that had gone noisy as a failure mode. Just my personal experience, but I find most reverb noise is coming from other things. ZIllions of amps like Fender and Peavey have driven reverbs with plain old 4558s for years.
meanwhile: 2N5089 or 2N5088? 2SC1815 always served me well and has your 1db noise figure.
I have played through many noisy reverbs, and offhand I can't think of many that had noisy designs. I often found ICs that had gone noisy as a failure mode. Just my personal experience, but I find most reverb noise is coming from other things. ZIllions of amps like Fender and Peavey have driven reverbs with plain old 4558s for years.
meanwhile: 2N5089 or 2N5088? 2SC1815 always served me well and has your 1db noise figure.
One that comes to mind is the Peavey Session/LTD 400. MANY many complaints about its noisy reverb.
Such as ??
I was referring to recovery circuits rather than drivers.
Yes, the 2N5089/5088 come in very well in my chart using the Horowitz-Hill table. The 2N3904 comes in as the best NPN TO-92 transistor, but I'm still trying to figure out the vast difference between the datasheet quoted noise figure and the Horowitz-Hill figures. 2SC1815 not listed. Datasheet noise figure of 2SC1815 is quoted on datasheet as 1db at 10K Ω.That's a cute way of hiding the truth, really. Calculates out to ~ 6.5 nV√Hz. Not particularly quiet.
Last edited:
The higher the impedance the less voltage noise matters and the more current noise matters.
Voltage noise is not the only noise (and often the least important source of noise in real circuits).
The first thing to establish what the source impedance actually is, so if the inductance isn't known, its difficult to judge - some measurements would be needed.
Voltage noise is not the only noise (and often the least important source of noise in real circuits).
The first thing to establish what the source impedance actually is, so if the inductance isn't known, its difficult to judge - some measurements would be needed.
Are you comparing them at the same collector current? The higher the collector current, the less equivalent input noise voltage and the more equivalent input noise current.
I looked up the On Semiconductor datasheet, https://www.onsemi.com/pub/Collateral/2N3903-D.PDF , and it specifies a maximum noise figure of 5 dB at 1 kHz for a 1 kohm source impedance, IC = 100 µA, VCE = 5 V at 25 degrees C ambient temperature. According to figure 10, the typical value is 3.6 dB under the same circumstances. Figure 9 shows that 1/f noise is not negligible at 1 kHz.
I think the best data you have for On Semiconductor 2N3904 transistors are those in figure 9 and figure 10. The 5 dB maximum spec is just a rejection limit set far enough above the normal noise level not to get too much yield loss. Still, 3.6 dB is a bit high.
It could very well be that the 1/f noise and base resistance noise vary between manufacturers. Apparently On Semiconductor's 2N3904 isn't as good as the ones measured by Horowitz and Hill, or whoever measured the data for figures 9 and 10 of the datasheet didn't measure very accurately.
Last edited:
I wonder why Horowitz-Hill didn't include it in their survey of Low-Noise Transistors?? It does have a very low datasheet spec for Noise Figure......
I believe it used to be the case that low (flicker) noise had to be screened for on a part-by-part basis, taking a lot of time (since flicker noise is intermittent), and thus putting up the cost significantly - thus a part designed for low noise back when flicker and 1/f noise were more prevalent (poorer processes, not using buried junctions?) would have a different code and cost noticably more.
For instance the NE5534A is/was a screened version of the NE5534.
It makes sense that any "general purpose" device with good typical noise performance isn't necessarily a good choice for manufacturing a product where noise performance is critical, even if its better on average than the device actually used(!). People can self-test components, but often its easier to pick a device with guaranteed adequate performance and pay for someone else to do the testing.
For instance the NE5534A is/was a screened version of the NE5534.
It makes sense that any "general purpose" device with good typical noise performance isn't necessarily a good choice for manufacturing a product where noise performance is critical, even if its better on average than the device actually used(!). People can self-test components, but often its easier to pick a device with guaranteed adequate performance and pay for someone else to do the testing.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.