Another 30+ year old precision analog trick of the trade many have forgotten and mentioned in Scott's 1983 paper linked above.
I do remember in my first year at Uni wondering why hifi amps didn't use the std instrumentation front end. It made so much sense to me. took me another 20 years to get cynical enough to realise why
You also get crazy precision in the RF value (do a Monte Carlo of 20 resistors in series using a normal distribution!) and a nice boost in power handing
Of course, I'm sure you know this. I haven't done a layout wth this strategy, but I do worry about parasitics.
Not least getting the pick and place to chose resistors next to each other on the reel. Can get you remarkably close. Also known for over 30 years
I'm not sure why that would be. I assume that the parts are not reeled in manufacturing sequence, but the reeling machine is filled with a "box", or bin of parts to be reeled. If the manu reels lot by lot, then there will be a break in the reel based on lot size.
As well, the films I've been using were lasered to tolerance. As such, the distribution will not be gaussian around the value. If the manu has a tighter tolerance spec'd part, there will be a hole in the main distribution. For example, there will be no .5% resistors in a 1% batch if the manu sells .5% resistors.
Also, the distribution should be one sided if the laser setup hits then measures. It will stop when it reaches the tolerance band, so the values will always be on the lower side.
That works well if there is no trimming involved in the production line.
John
I don't know how the reeling is done or if this was an old wives tale passed on, but I shall try and find more. Certainly you are likely to get clusters off the same slab (biscuit, whatever) next to each other unless the stir them.
And if the 0.05% parts are selected first, then the 0.1% then the 1% you are right about the distribution, which suggests 1% parts, whilst they might not be the exact value you wanted might actually be very close to each other.
I assume we can agree that, given process drift, parts made the same week will likely be in a cluster vs parts made a year apart?
Why then do you drive 20 year old Acura and complain about DAC's you can't afford?
If they pull them out of a bin, then the 1%'rs will be a two lobe asymmetrical distribution about nominal, a gaussian distribution with a big half percent hole in the middle. (half percent on either side of nominal.)
Absolutely.
John
I'd been taking your comment on laser trimming giving asymmetric distribution as meaning you'd get tight clustering on one side? Did I misunderstand?
Like that guy a few weeks ago complaining that his 10% PSU caps were all around -5% and he wanted to return them
It's not going to matter how precise the input resistors are if the source and cable aren't considered. I think it was Richard Marsh who pointed out in an Audio Amateur article that the entire chain - preamp output, cable, and input of the amplifier have to be tuned for optimum CMR and a variable C for HF tuning would be useful. Changing the cable or preamp would detune the CMR trim.
Hence the whitlock designs THAT makes and why I like them so much.
But yes, if you want balanced (note balanced not necessarily differential) you need to do your whole chain. Not that much of a bind for the DIY type.
No, you didn't. If they simply lasered them to a nominal and dropped them into a bin for later sort, then the lasering process will still produce a symmetrical distribution around a nominal. Selection would be from that distribution.
If they simply lasered to the exact tolerance (stopping once tolerance is met), it will be lopsided on the low side.
Yup. Very hard adding material back on with a laser.
Actually, not so much so anymore. I watched a youtuber where GE laser sinters engine parts (I think they were fuel mix distribution nozzles) that cannot be made any other way. Really really cool.
Then, about a week later, listened to a talk on site by some fusion guys who want to make the entire coolant loop plumbing (pump with impellers and all) via laser sintering that drops into the center of the tokamak. About one meter cubed.
Next, some diy'er will be laser sintering copper directly onto thick build anodized alumina sheet..then ruthenium resistive material, then yes, sintered metal films...even aluminum and tantalum, which can then be oxidized to make capacitors..then transistors soldered on and wire bonded.. DIY Power hybrids..
Then, sintered 3-d structures for jacks and selector switches..
Remember, you heard it here foist.
John
Ah, makes sense. I haven't seen laser trimming in operation since the mid 80s but that was veeeery slow and for some military application with vernier trimming. Assume on thin film production it's impressively fast. off to youtube...
Yes, done in bulk and very fast! The speed / cost depends on the cut type. A plunge or L cut where a straight or "L" shaped nick is made to one side is cheapest, but the best performance is where the printed film is equally thinned from both sides - needs to be done this way for RF applications, for example - but that's slower so costs more...
I really see using 2% resistors instead of 1% resistors almost silly. 1% is the new standard, just like 5% was standard 50 years ago.
It used to be that a single 1% resistor cost as much as 7 high beta npn transistors when I was at Ampex, almost 50 years ago. We did not use them much. 5% Allen-Bradley baked carbon was the standard. Today, we can't find anything much but 1% resistors.
When, in the late 70's I got access to military surplus resistors, I got the whole range: .1%, or even better, sometimes. Some resistors were really well made and actually beautiful with glass enclosed casings and gold plated leads. I still have many hundreds of them left. However, when I made a preamp with them, the preamp did not sound very good. I finally decided to avoid the 'pretty' resistors, because they seemed to create disappointing sound when used in feedback loops with familiar, and proven electronics, like Levinson modules. I attribute this to magnetic leads and perhaps other factors.
When it comes to precision, the RIAA network is one of the most critical that I have to deal with. For the Vendetta Research SCP-2 phono stage (one of my very best) I settled for 1/2%, because they were available, HOLCO, (because they were known to sound good, AND they had thicker leads than normal 1% resistors). Then of course, CAP tolerance becomes critical, so I special ordered a batch of 1% polystyrene capacitors, that I individually screened in groups, so that the L and R channel would have the same RIAA, as best as I could easily make it. I am not sorry for my choices 30 years ago, and I would do it again today, if I could.
I think that lowering standards just to use up what you have laying around your lab, is a cop-out.
It used to be that a single 1% resistor cost as much as 7 high beta npn transistors when I was at Ampex, almost 50 years ago. We did not use them much. 5% Allen-Bradley baked carbon was the standard. Today, we can't find anything much but 1% resistors.
When, in the late 70's I got access to military surplus resistors, I got the whole range: .1%, or even better, sometimes. Some resistors were really well made and actually beautiful with glass enclosed casings and gold plated leads. I still have many hundreds of them left. However, when I made a preamp with them, the preamp did not sound very good. I finally decided to avoid the 'pretty' resistors, because they seemed to create disappointing sound when used in feedback loops with familiar, and proven electronics, like Levinson modules. I attribute this to magnetic leads and perhaps other factors.
When it comes to precision, the RIAA network is one of the most critical that I have to deal with. For the Vendetta Research SCP-2 phono stage (one of my very best) I settled for 1/2%, because they were available, HOLCO, (because they were known to sound good, AND they had thicker leads than normal 1% resistors). Then of course, CAP tolerance becomes critical, so I special ordered a batch of 1% polystyrene capacitors, that I individually screened in groups, so that the L and R channel would have the same RIAA, as best as I could easily make it. I am not sorry for my choices 30 years ago, and I would do it again today, if I could.
I think that lowering standards just to use up what you have laying around your lab, is a cop-out.
Attachments
feedback divider resistors
A lot of this stuff is prior art, including Whitlock's a.c. bootstrapping per se. I haven't inspected the THAT part's patents closely, but I do recall being told that a great deal of their work leading up to the release of the Ingenius part pertained to making the parts robust against various overload conditions. The arguments presented for how noise is contributed by high-value input termination resistors are mostly specious, as the much lower source impedance will "short out" the resistor noise.
Yes, this is key, as it indeed removes almost all thermals and voltage coefficient modulation. Short of that, the tempcos available from thin films are quite small.
A lot of this stuff is prior art, including Whitlock's a.c. bootstrapping per se. I haven't inspected the THAT part's patents closely, but I do recall being told that a great deal of their work leading up to the release of the Ingenius part pertained to making the parts robust against various overload conditions. The arguments presented for how noise is contributed by high-value input termination resistors are mostly specious, as the much lower source impedance will "short out" the resistor noise.
- Status
- Not open for further replies.