how to determine cutting head max power?

[himynameisjacob]

Is there any way to know what is the max power of the old mono cutting head? I mean, max power RMS I can send before I fry it? I want to build a power amp for Neumann MS52H, but don't know where +/- to start.

Thanks!

[boogievan]

Your cutting head needs a lot more power than it can sustain for even one full second in order to make good sound. Al Grundy explained that the average cutting head can dissipate the heat of only 6 Joules (which is 6 Watt-seconds). But it needs amplification of high frequencies for fractions of a second, during transients (cymbals, claps, claves, etc...) sufficient to produce stylus accelerations in excess of 1,000's of G's. Therefore, your cutter needs signals from an amp of at least 250 W, and, ideally, closer to 500 or more, so that you'll have reserve power - higher than that which is sustained - for very brief, high frequency bursts (whereas hf reverb tails - being 'post-decay sustains' (as in ADSR) - involve much less velocity than their 'attacks').

The RIAA filter (which requires treble to be cut much louder than midrange) and close-mic techniques and The Loudness War conspire against the glue that holds the drive coil together, but good sound is king, and cutting with peaks no higher than 0 VU (ref: 1 kHz @ 7 cm/sec peak lateral velocity) on pickup (i.e., playing your 1-kHz nominal-level cut causes your turntable cart to deflect the VU meter indicator the same amount as when playing back your pressed test record''s cut of 1 kHz, lateral) should be safe - even if your amp can make 1,000 Watts, as long as you use a de-esser / acceleration-limiter in the progam path and lean towards a symphonic (acoustic, natural) frequency curve with your mastering processor settings.

Recently, others have been discussing using a low-resistance, fast-blow fuse in series with a Preseto cutter. But you need to pick a fuse you can use, innit? Ideally, your amp will have a circuit that constantly measures the DC resistance of the drive coil so that, as soon as the resistance of the coil that was measured, cold, doubles, then it must be too hot (since heat makes metal's resistance get higher) and the amp will automatcially mute the signal that's been heating the coil (and the glue). Worth implementing and sounds as if markrob can help you, from what he recently has posted...

- boogïe

[himynameisjacob]

Thanks for that detailed reply. VR uses 1A fuses inline for the each channel of c-head. 2000W amp as far as I remember (?). But each head is different. Presto K's used 2x6V6 (or 6L6) tube amps, and it was just a hair quieter than VT, and those amps would make around 30w clean so I am confused... Isn't it the best way to try to obtain "undistorted" sound starting from low volume, and raising 1-2dB each session, to get close to VR cut loudness? I am aware that amp in my lathe is shot so need to design something that will do the job (not tube, not class D as I read it's not good - right?)

[boogievan]

What's not intuitive is that it takes very little power to cut around the primary resonance of the transducer which is usually around 1 - 2 kHz. There's a slight increase in power required for low frequency signals, as one selects a lower and lower test tone, but only up to about 5 Watts - certainly much fewer than 30.

The high power amplification is for high fidelity in the third octave. It's not going to increase the overall perceived loudness, much, because that's judged most heavily around the easist frequencies to cut (...and to get across a telephone line for that matter) - namely, midrange, which is mostly the second octave. Whereas the details, realness, intimacy, excitement, etc..., added by a flat frequency response (after de-RIAA) above 10 kHz is where the power needs to be above 250 W, and, yet, even 30 W can be way too much, if prolonged - as in - for more than a fraction of a second

Rather than over-limit current with a fuse that's intolerant and fast-acting, I recommend you implement a (sonically-blameless) heat-protection circuit based on the DC resistane of the cutter's drive coil. When it's cold, it will measure something, such as, say, 4 Ohms. The cutter should probably have a Mercury-wetted reed relay cut off the Program Drive signal as soon as that coil measures, say, 8 Ohms. If your cutter measures 8 Ohms across the drive coil contacts when cold (i.e., at room temp and not signaling), then it can probably take the heat that makes it measure 16 Ohms during a cut before it needs to be disconnected due to having overheated.

{The Mercury in the relay will protect the signal bypass of Cutter Off from bouncing back to On as the relay shorts, since the fluid metal will instantly keep its contact as soon as it's made, even if the contactor slightly ricochets on impact.}

- bøøgie

[dmills]

Loose the old mercury relay, we got better ways to skin that now.

Skyworks have an isolated gate driver that is made for switching a pair of back to back mosfets, and will do it in microseconds that is just ideal for speaker protection applications, which is really what we are discussing here, cheaper, and no mechanical limitations (Mercury relays are sensitive to orientation).

Do fit a 120V or so bidirectional transient suppressor across the output to prevent the inductive spike from disconnecting the head under load damaging anything (Easier then that mess of caps and diodes in the Neumann).

The old Neumann circuit breaker is easy to replicate, especially as we now have small modular isolated DC-DC modules to provide the floating supply required (All the doings float on the bridge output), but the power amp will need to have a way to dial in a small dc offset to provide the sense bias, note that this design assumes a feedback head and has the phase lead network as part of the bridge circuit.

Incidentally, that +40dB at high frequency (Actually it is not quite that bad due to the velocity and acceleration limiting) means that you need to have a bit of a think about the dimensioning of the RC network across the amp output that is usually required for stability, that resistor can get rather hotter then is normally the case.

There are some very interesting AES papers on the use of an ultrasonic signal for temperature measurement that gets a significantly faster response then the rather laggy DC based approach which relies as much on scaling the current trip point with temperature as being a pure temperature trip.

[boogievan]

Just because something's new, doesn't mean it's better, so, I'd not 'lose' that (single) Mercury-wetted reed relay (per channel), nor would I add MOSFETS to an existing circuit in order to replace the Merc. (It will have an arrow pointing up, and you'll just have to remember not to turn on the amp unless the chassis is horizontal, innit?);

We don't want the signal to go through more semiconductors than necessary. (That's why we insist on single-ended amplification in the mastering chain - less is more (open))

Freddie Mercury's mechanical switch - as long as there's only one per channel (as Sherwood Sax strongly suggested) - will keep your safe switching 'audiophile'. Semiconductors introduce a coldness via stochastic noise that is best minimized for implementation, and none of the transistors in the Ortofon amps, for that matter, are MOSFETS, so Luke Skywalker would have to kludge them on.

From sound-au.com:
'When MOSFETs fail, they almost invariably fail short-circuit (like most semiconductors), and it is conceivable that a failure could go entirely unnoticed until your speaker catches on fire. It is essential to make sure that failure is rendered highly unlikely, or that some kind of test process be incorporated (which adds further complexity of course). Quite obviously, a conventional relay can fail too, but they are generally extremely reliable and have no sensitive electronic bits in them.'

& Jenny List 'gets' it: https://hackaday.com/2018/03/10/only-mechanical-relays-will-do-for-automated-hi-fi-audio-source-switching/

Please don't add an ultrasonic signal to a cutting head's coils... Difference tones may fold into the high frequency range where feedback phase twists positive and result in oscillation. Fewer powered components is the safer way, and the sound of one Mercury relay (per channel) is the sound of gramophone cuts. See you down in Davy Jones's locker, Jimmy Page †

- boogie

[dmills]

The noise of the semiconductors is all too often dwarfed by the noise of the resistors! You have to be very careful if you want the noise of a decent opamp to dominate over the Johnson noise of the resistors that surround it. Possible to bollox it? Sure, anyone can screw it up, and most designers will occasionally if being honest. A single 10k resistor into a high Z input is noisier then a humble 5532 fed from a suitable impedance!

Usually harsh, cold sounding audio from a semiconductor thing has me reaching for the scope cranked out to see up to at least 10MHz, amazing how often it turns out to be something honking, usually where someone has used their 'golden screwdriver' on the choice of opamps... A retrofitted LM4562 with grossly inadequate decoupling or excessive capacitive loading, looking at you! The other fun one is gear with WAY too wide bandwidth, there is nothing good about an audio design that extends out past 100kHz, it just begs for RFI, this is one place where a jfet part is usually superior, less prone to demodulate the local AM.

Not that noise is much of an issue in this application (Exceptions: feedback preamp, and to some extent 1/f post IRIAA stage and in the feedback preamp, to some extent power amp feedback path), I mean the ultimate medium is going to manage 60dB (Some say 80dB, never seen it), and decent analogue has about 120dB of dynamic range, so even if the RIAA brings the bottom end up by about 20dB on playback (And 1/f brings it up by 20dB on cutting) you are still not going to hear it.

Mercury reed relays opening under significant power are actually surprisingly unreliable, and mercury DISPLACEMENT relays (Which are the normal choice for a power relay when mercury is desired) are all kinds of slow.

Given a pair of modern 200V power mos back to back with a total of maybe 0.05R on state resistance, and a constant gate source voltage (So not acting as an amplifier) I would consider that the 50+ amp 10us SOA at a few volts is sufficient that the amp current limit is going to protect everything for long enough for the over current trip to activate and switch off the mosfets (in microseconds, not the milliseconds that a mercury relay would take). In this application, the mosfets are NOT amplifiers and in fact operate deep in the ohmic region. You can obviously screw this up as well, but it is possible to do it right.

You can organise a self test at power on by ensuring the bridge voltage is acceptably close to zero before switching the fets, and this sort of thing is cheap enough that both poles can be switched so no single failure will screw it up (And so you can use a bridged amp if desired).

Finally, remember that the whole bridge and switching doings are INSIDE the cutter head feedback loop (As they must be to flatten the velocity response given the phase lead network for the feedback that is usually part of the bridge at DC), so any distortion here will be very much reduced by that loop.

Mercury relays work obviously, and particularly if refitting an existing cutting amp, keeping it close to period is a valid approach, but there are other ways to skin it.

Yea, I have a cutting amp design on the blocks, 350W or so each side, (12A @ 60V peak before the amp current limit kicks in (Should never happen, it is a protect the amp thing and only has to hold the outputs within SOA for long enough for the output disconnect to trip on over current), should do for my Caruso), and yea there is a mosfet based disconnect that (horror of horrors) uses a toy microprocessor to run a rather better thermal model (integrating I^2dt is easy on a computer, and with the measured head temperature as the other input, wonders become possible), the thing kills quite a few sacred cows, but it will be fun to test. Class D was briefly considered but to get the bandwidth wide enough to not screw with the phase margin of the loop I would have needed to go down the GaN rabbit hole....

I have done the mercury reed relay dance (In a transfer console design), pain in the arse it was, but that thing is SILENT and switches cleanly. Not sure I would do it again, pain in the arse to build, and ties up far too much rack space.

[boogievan]

It's not noise that semiconductors make I don't like, but the enharmonic distortion. I've listened to the same snare drum transient as portrayed by several popular op amps and they all add a certain low level screech to the decay of the snares - making the air-band false.

Worse, failing closed is supposedly the more likely doom for the ssr - so you'd still have to add a mechanical relay as a safety? (abducted emoticon);

[dmills]

Are you sure those opamps were properly stable?
I have heard some weird things from such (You always have to listen, it is a critical part of testing a design), but it always seems to come down to instability or running the thing into clipping (Recovery from this is often NASTY, just a trade off with using lots of NFB). The (excellent) LM4562 is one of my usual suspects here, it is FAST, but does not at all like capacitive loads, and needs RF layout to be really stable, worse folks keep replacing 5532s with them without considering the decoupling or the speed, then crying about the sound of the opamp! I would suspect clipping here, snare being what it is.

Incidentally, it is worth comparing a snare recorded from two different distances with condenser mics, turns out the condensers sometimes have a really nasty intermod problem up in the low ultrasonic that no one wants to talk about.

Fail closed is generally the fault mode in a semiconductor right enough, but there are reliable ways around that (crowbar the amplifier power supply to take the fuses, or double pole switching with fault detection, not a big deal).

You can also get stupidly butch switching mosfets these days, and between a fairly crude current limit in the power amp (A diode clamp to limit the voltage across the emitter resistors) and an absolutely massive peak current rating on the switching fets, I am not too worried about eating the mosfets. Don't forget that there is usually a 4R or so resistor shunted by the phase lead cap in the bridge circuit, so pathological DC current is more limited then you might expect (Still enough to kill the head quick like, but the protection network has a defined short circuit current to handle even if the amp completely shits the bed).

[boogievan]

Yes, the op amps were stable. There was no clipping - plenty of headroom. At first they, each, sound good, detailed, even exciting. But very close A/B-referencing to the original signal revealed sound-trails being slightly changed - very slightly - meaning, just in the sustain and release modes (re: ADSR), where I think our subconscious processing is keen and which suffers from the listening fatigue introduced so often by digital audio listening sessions, for that matter .

The op amp that won the shootout was the OPA213P with only an 8 MHz bandwidth. It's used by Dangerous Music in the Bax, btw. I believe its advantage is that it has a very fast settling time of 0,7 µs. Whereas, the LM4562 has a relatively sluggish 1,2 µs settling time. (nearly twice as slow). The Burr-Brown's slew rate is identical to the 4562, and OPA's THD+N of 0.00008 % is already vanishingly low (even if technically twice as great as the reported 0.00003 % of the Linear Monolithic op amp).

Going up above 10 MHz is for coms rather than 'audio', ime, reminiscent of how Dan Lavry famously objected to driving digital audio converters above 2x sample rates - the settling time was not adequate for best precision - especially in the low frequency reconstruction - and that 60 kHz F/s would be an ideal sample rate for best sounding audio, up to a Golden Ear's 30-kHz requirement.

[dmills]

That's jfet input isn't it?

Probably better from an RFI perspective if the inputs lack solid filtering, but the noise impedance is also going to be way higher then the bipolar parts, so it will want a different feedback network for best performance.
I certainly don't want to be designing audio that passes the AM broadcast band, I mean have you heard the American AM talk stations, utter cringe (Even if you are a republican) and the last thing I want blowing thru my audio doings.

My beef with this sort of test is that it tends to treat the opamp as being the thing, when you really need to be looking at the opamp AND its surrounding components and layout, never the opamp in isolation, in my view an opamp is a component of a circuit and it is the performance of the complete circuit that matters and that is mostly a topology thing.

Have you by any chance played with that conditionally stable composite opamp that Samuel Groner designed? I keep meaning to, but too many projects, you know how it is.

Remember always that GBP is Not the bandwidth of the signal you are working with, it is where the amp runs out of open loop gain, and you really want loads of open loop gain up to the point well above where you have lowpassed the signal, this is key to avoiding the risk of intermod due to device non linearities, you must still have open loop gain sufficient to ensure the feedback is sufficient to linearize the thing up to well above the highest frequency present (And not just what you can hear).

My general philosophy (for what it is worth) is to lowpass at the input with about a second or third order filter at about 100kHz which is a compromise between minimising phase shift in the audio band and keeping the radio stations out, MFB filters are better here then Sallen & Key because they don't tend to come back up when the amp runs out of GBP. I like a passive first stage if I can, and the inverting nature of the MFB means that Mr Pultzeys clever ground sensing trick to make the signals independent of the ground plane current comes naturally.

There are multiple approaches to this stuff, and I readily admit to having listened to far too much rock and club stuff in my youth to be able to claim golden ears now (50 years old, my days of hearing the CRT line scan are behind me), so I am not in the habit of telling someone who can hear something that it is not there just because I cannot hear it, but you can only do what you can do.

IIRC Lavry wanting 60Khz SR was in the days before oversampling filters where CD players used analog reconstruction filters in the 20k - 22.05kHz gap, with the obvious implications for ripple and phase weirdness, opening the transition band out to 12kHz would have made the compromises there FAR more reasonable. It would still help with the modern use of half band filters that produce a narrow band of aliasing, but we probably don't NEED it to sort that out.

Anyway, we are WAY off topic!

[boogievan]

Am using these op amps at unity gain in the test circuit. So the bog-bandwidth used by Max Powers, in this case being 1 MHz, is irrelevant. All the more, this is overkill for audio.

Lavry did revise his work-refusal (re: manufacturing 4xFs converters),

Please note, however, I'm not talking about the deleterious effects on group delay in the third octave caused by the steep, anti-aliansing filter, which is the infamous, though subtle frown-face in digial audio's air band that occurs in... / is still revealed by ...all 2x-upsampling DACs, most notably when switching between 44.1k and 48k Fs, having the same analog source being digitized(, with gains realized less significantly, respectively, when switching up 88.2k and 96k Fs....). We're talking, instead, about the settling time of operational amplifiers (...when being run in oversampling converters!) and their accuracy (saved as many memories) at speed and that ironically affecting the low frequency sampling measurements, down in the first octave. So, not the filtering issue - a mechanically-based digitizing issue.

- boogie

[dmills]

I think the RF present in the output of (particularly NOS) R2R DACs when combined with the bipolar inputs on something like a 5532 sometimes has rather 'unfortunate' effects, mostly down to slew rate that possibly explains some of the dislike of that part in this application, as well as why such DACs are thought to sound different. This IMHO is one of those cases of different being mistaken for better.

This however is once again a topology thing, and is largely audio folks not understanding that a DAC (including to some extent its output) is an RF thing as much as it is an audio one and should be treated as something outputting fast edges, particularly a current output DAC feeding an opamp based I/V stage needs to slew like an absolute bastard to avoid banging the current output into the limited compliance voltage of the DAC until the opamp catches up. Something to be said for a fast common base input section there!

The interesting thing right up against Fs/2 on almost all modern over sampling DACs is that because they almost all use half band filters (Because cheap and minimal silicon area!) in the interpolators, they are typically only -6dB at Fs/2, falling like the side of a house of course, but there is a small region just below Fs/2 that is a mess of aliasing.
The cure is a properly designed digital filter preceding the DAC to take the top few kHz off.

This is getting stupidly off topic!