EEVblog #529 - HP 35660A DSA Upgrade Investigation

A teardown of the front end PCB in the HP 35660A Dynamic Signal Analyser.
Dave does some preliminary investigation to see what opamps and FET's are unsed in the analog front end to see if they can be upgraded for better performance.
This is more for Dave's record than interesting info.
Datasheets:

EEVblog Main Web Site:
EEVblog Amazon Store:
Donations:
Projects:
Electronics Info Wiki:

[bruhnstv]

Dave, you might note that the power to the dual FET is bootstrapped to follow the input voltage. There are some implications there to why you see so much noise. It's not a BNC: it's an SMB! Typical HP part numbers: 1826-xxxx are analog ICs, 1820-xxxx are digital. So the 26-0715 really is 1826-0715. Sometimes you can get lucky with a google search or similar, as you can with this one. If you get really lucky, you score an HP cross-reference booklet. (1853=PNP, 1854=NPN, 1855=FET)

[EEVblog]

Silly me, yes. It connects the grounds together. I should have done more than glance at the schematic.

[Zauviir]

I love this older through hole stuff, awesome to see what went into it!

[peterpv]

You're correct if the FET stage would have sufficient gain. But this circuit looks like a voltage buffer to me so gain = 1 so the opamp will most likely be the largest noise contributer.

[WhitentonMike]

Dude, you nailed it! Take a look at 4:51. There is a shield that runs right down that line. They didn't want the shield to wear through to the traces so they ran all of them on the back of the board. I didn't even notice it during the disassembly. Dave was talking about the BNCs and I wasn't paying attention to the shield. Good call on that one.

[stephenbalogh]

9:42 This is not only electronics, this is a work of art. So much beauty in the components and construction in this view. I love the old school boards.

[bruhnstv]

Seems like an easier way to get lower noise (and what I did, in fact) is put a low-noise ext preamp in front of it. You can even use the HP35660/35665/35670 to measure the gain of the preamp as a function of frequency, store that in a register, and use the math functions to more directly display calibrated readings. My application was for a low-impedance source; I used an AD797, 2 9V batteries for pwr, and well shielded, to keep the preamp noise very low. For AC cplg: 10uF of polyprop caps.

[gilgameshismist]

Love to see you speaking in the reflection of the screen!

[philpem]

Quick followup - those transistors are HP 1855-0460. The general rule with HP IC part numbers is, they drop the "18" if there isn't enough package space to print it. I tried to look it up in the HP_Agilent_Equipment archive but the horrible new Yahoo Groups interface is frustrating my every attempt!

[PaulHartyanszky]

3: A hybrid BJT and op amp stage. Cheap options include using 2n4403 in paralell as is very common for high end moving coil and balanced mic amplifiers. The analog devices sssm2220 PNP match array has a .32nV rt Hz amplifier circuit design in the application notes (fig 16). It parallels 3 arrays ($8 each) and uses a similarly priced ad8671 as well as .01% resistors. Expensive, but components are readily availible.

[SwitchingPower]

Dave the HP 1855-0460 matched FET on the input is a 2N5545 "Monolithic N-Channel JFET Duals"

[KX]

AD8675 looks like a nice part, which I had previously overlooked. Quite similar to the OPA209 I've used before. Cheers.

[RobB_VKES]

I agree. the FETs ought to bring the signal level to well above the noise floor of the op amps. Probably the logicHP put in initially.

[PaulHartyanszky]

I've been interested recently in design of amplifiers for noise measurement and microphone preamplification - though the latter has low distorsion requirements as well. For highest performance, I think your options are:

1: A low noise op amp such as lme49990 (.88nV rt/Hz), AD797 (.9nV) or MAX9632. I have a stack of lme49990. They are much cheaper than ad797, have lower 10hz noise and has ridiculously low distortion. However, it does higher current noise so beware of high source impedences.

[bruhnstv]

Yes, you're right: the 4 FETs do contribute about 20nV/rtHz at 1kHz, and that would be improved by using better FETs. Nonetheless, you'll quickly get to a point where noise is dominated by the output stage of the input buffer op amps, because the input buffer stage is run at unity gain. And of course you'll also be limited by the noise in those 5k resistors.

[bruhnstv]

I should add: the 2N5545 is indeed not particularly low noise, but it's lower noise than Dave's seeing. You'd gain a _little_ by replacing it with a more modern, lower-noise JFET pair, but you'd be disappointed that you didn't get down into the low single-digit nV/rtHz spec'd for your part. You probably WOULD improve the noise at very low frequencies significantly; the 2N5545 has a pretty high 1/f noise corner.

[KX]

It's black. At and around 10:47 you can see black text around the board, and the black line down the middle in question branches off to the right clearly at 11:05 crosing several top layer traces. So the line must be silkscreen dividing areas of the board for some reason.

Dave, the problem is, more modern op-amps than the AD797 usually only have SO-8 packaging instead of DIP-8. If you can upgrade the PCB using SO-8 op-amps, you could try the LME49990 from TI, it has better than great specs, and it doesn't need compensation as far as I remember.

[philpem]

"26-0715" --> HP part number 1826-0715. Google for a HP part number cross reference an look it up and... it's an NE5534:
1826-0715 2910 NE5534AN
2910 is the manufacturer code for Signetics.

[PaulHartyanszky]

2: Design a special low noise op amp out of discrete components. This is difficult, potentially expensive and other parameters are going to suffer. The lowest noise I've seen online is Samual Groner's SGA-LNA-1 precisely for this application - 0.5nV rt Hz. It uses hard to find transistors, however.