Cheap as Chips OPA1688 Low-THD Muscle Amp

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Only design decision based on sim was to use 1688 for the voltage gain driver in order to get a 2nd harmonic dominant profile (wheres 49720 was third order dominant). And low ESR rail caps to get low distortion based on low Lser values in sims. So is the 1688 model not accurately showing relationship between PSU series resistance and distortion? I think the model of the 1688 does account for distortion. Lack of an explicit section on THD doesn’t mean it doesn’t model distortion arising from other non-linear effects as we can see when changing the Lseries or the load impedance.

Please show me an opamp model that explicitly accounts for THD and has “transistors” inside.

Saying that a Spice model of an opamp is useless because it doesn’t contain transistors is like saying a model of a transistor is useless because it doesn’t explicitly contain a section of N and P doping on valence electrons.

Now you are making wrong generalization. I said that modeling distortion intrinsic to the opamp with the given model is useless, I didn't say the model is useless. Just reread what I have written earlier.
 
Last edited:
Please show me an opamp model that explicitly accounts for THD and has “transistors” inside.

To my knowledge, the device-based models went by the wayside around the same time the various manufacturers stopped publishing the equivalent schematic for their ICs. The models available today seem to do a good (but not perfect!) job at capturing the DC and AC behaviour. In some cases (this is pretty rare) they capture transient oddities as well.

I'm not aware of any model - certainly not any model of a modern part - which captures THD or "has transistors inside" (= device-based model). And that's been my central point all along: Your simulations show that the amp is not fundamentally broken, but does NOT capture the THD or the harmonic structure of the distortion. Therefore, chasing a certain harmonic structure or counting the zeros on the THD figure based on the simulations is an exercise in futility.

Saying that a Spice model of an opamp is useless because it doesn’t contain transistors is like saying a model of a transistor is useless because it doesn’t explicitly contain a section of N and P doping on valence electrons.

I'm not saying that the model is useless. I am saying that it does not capture THD (and based on your more recent statements, you seem to agree on this). As far as device models go: There are at least two types of models: Physics-based models, which use the doping concentrations and doping profiles to model the devices. And behavioural models, which start with measurements of the device behaviour in the lab. Various parameters are then derived from the device curves and used in generic model equations. SPICE models fall into this category. This type of models are valid for the range where the devices were measured and under the conditions they were measured. If you use a device in a mode where it has not been characterized (sub-threshold for a MOS, for example), your results will likely be utter garbage as the model is not valid there. Just as your THD results are meaningless because the THD isn't modelled.

Tom
 
Last edited:
Why go all defensive on this xrk971? The limitations of opamps models are well known wrt thd. And simulation time or response to supply impedance are not good indicators of the contrary. Even the distortion spectrum is suspect so I wouldn't make design decisions based on that.

On a more constructive note, you might want to have some gain in those output buffers. Hunt down jcx posts on multiloop, he posted abundantly on that.
 
Founder of XSA-Labs
Joined 2012
Paid Member
I don’t need any gain in the buffers as the voltage amp driver is providing that.

Defensive? I think I said that I agree that sims are optimistic. I just take issue when people summarily dismiss a commercially produced model by major manufacturer such as TI as useless. It’s quite an arrogant attitude and uncalled for. As Tomchr pointed out, the model of the 1688 is actually pretty good - just lacks specific section on THD. I will build this and we will see how it measures.

Let’s move on.
 
You are ego-tripping, Shaq. I think X is publishing his HPA for the benefit of others; you are publishing to demonstrate how smart you are. Disgraceful behaviour.

I looked over your circuit. Nice circuit, competently designed. But no idea from the results you published how it actually sounds, and if you could live with it. In the experience of many, THD is only one small aspect of the assessment of any amplifier.
 
xrk971: noone is saying the model is useless. Just warning you on its limitations. Wouldn't you agree Tom is pretty well placed for that, having worked for TI ? TI released pretty poor models in the past (lm3886...) and never bothered to correct them so I wouldn't trust them just on their good name too.

As for gain in the output buffers, it's not a matter of extra voltage gain. It's a matter of further reducing distortion. Just search for jcx posts; he explains it much better than I could.

Edit: the wson8 package is cool... Pretty much impossible to solder for a diyer without risking damaging the opamps. Do you plan to sell PCB with pre soldered ICs ?
 
Last edited:
My 2c as someone working in this field. Model if you must, perhaps just to ensure your idea will work. With time, experience can be enough, and you can skip the modelling and get a long way with an understanding of the devices, your goals, and Ohms law. Then build it and measure it, and do your fine tuning from there.
 
I forgot the major benefit of having gain in the output buffer. With only a gain of 2 at the output, the input opamp can run with 2/3 of the rails voltage (taking losses into account). It means you can regulate the rails for it down from the power rails. Since the accuracy of the amp depends mostly from the input opamp, you can then use the best regulator for it and a simple one for the outputs (or even none). In a high power design, it cuts costs and dissipation by quite a bit.
 
Founder of XSA-Labs
Joined 2012
Paid Member
the wson8 package is cool... Pretty much impossible to solder for a diyer without risking damaging the opamps. Do you plan to sell PCB with pre soldered ICs ?

It’s not for the beginner, but with solder paste and a hot plate and skillet, can be done with practice. I may offer pre-populated SMT parts - depends on level of interest. I considered easier to solder SOIC8 package for the buffers but they have no thermal pad. At max current of 75mA, that’s 27mA rms and +/-18v rails, that’s 1w per device. This board will actually have stick-on heatsinks on the bottom side where the vias for the thermal pads come out.

Total maximum current draw for 8 buffers is 1.2amps for both channels. I will just use a 2amp capable power supply and regulators.

The use of low gain 2x on buffers to relieve voltage rail requirement on driver is an interesting idea. Seems like adding nested feedback - if you have links, please share.
 
Last edited:
Those models are not commercial products. You didn't pay for them.

I think we're splitting hairs here. You're right that the models aren't a commercial product, but the OPA1688 certainly is. I think X is right to assume that a commercial product is backed up by a commercial-quality model, even if the model is provided for free. Sadly, that assumption is not valid for all parameters of the OPA1688 (or any other opamp) as you touch on in your following sentence.

Regarding its accuracy- you didn't even bother to create a test bench to verify parameters. How could you even tell if it is accurate if you didn't check it?

That should indeed be Step #1 in any simulation. Validate the model. Figure out if it provides the data sheet performance for the parameters you're interested in. If it doesn't, how far off is it?

xrk971: noone is saying the model is useless. Just warning you on its limitations. Wouldn't you agree Tom is pretty well placed for that, having worked for TI ? TI released pretty poor models in the past (lm3886...) and never bothered to correct them so I wouldn't trust them just on their good name too.

I actually disagree on the LM3886 model. It's pretty decent ... if you use it within its limitations. It gets the AC and DC parameters mostly right. It does not handle anything power supply related (infinite PSRR and supply current stuck at 50 mA regardless of load current) or anything related to clipping (so none of the buzz or other quasi-stability issues that need to be dealt with). But for loop gain analysis, it's a perfectly usable model. Just don't trust it beyond a few MHz and don't trust it when the output is near the supply rail.

That said, I do agree that the LM3886 model could use a lot of improvement, but for 80-90% of the design work, it's really quite useful. You'll just have to get used to the idea of having to prototype the last 10-20% of the design in the lab.

Edit: the wson8 package is cool... Pretty much impossible to solder for a diyer without risking damaging the opamps. Do you plan to sell PCB with pre soldered ICs ?

As I pointed out earlier: Watch the dissipation on those. If you're pushing them to the power dissipation limit (or even close) you'll definitely want to use the recommended footprint with thermal vias connecting the DAP to a solid plane.

I do agree that they're wicked cool, though. Automated assembly opens up a whole other can of worms. If you find a good assembly house, you'll likely find the automation to be well worth the expense. That said, it is a significant expense.

SOIC-08 can be soldered by hand, so if you'd rather avoid the automation can-o-worms, that's a very valid DIY option. Do beware, though, that hand-soldering of SMD passives often results in microscopic cracks in the components. That can lead to some really interesting issues (usually higher THD - including audible distortion). Just something to think about.

Tom
 
Last edited:
Tom: iirc, you mostly use Tina-TI ? I might be prejudiced against the on 3886 model as I am using ltspice and the pspice model is truly hard to work with, creating all sort of hangups. But I must admit it was made for pspice, not ltspice. The other models from TI are much better behaved under that respect. End of OT.

Xrk: I'll try to dig up a few posts later on but I'm in the mountain for a week with a truly atrocious connection. I'll try to find a spot with real 4g.
 
Founder of XSA-Labs
Joined 2012
Paid Member
If you do choose to offer an SMD pre-populated board, will that be automated assembly or frying pan DIY assembly by you?

Tom

Probably by me: but I use frypan to preheat PCB to 140C and then hot air gun from above to actually melt the paste to liquid metal. Seems to work pretty well for the few cases I have done with 20pin quad flat pack no lead TPS7Axxx.
 
Tom: iirc, you mostly use Tina-TI ?

If I use LT parts, I tend to use LTspice. If I use TI parts, I tend to use TINA-TI. In the past I've used the original text-based SPICE, PSpice, and spectre. These days I mostly use TINA-TI. All the simulators I've used have had their shortcomings and oddities. Different simulator, different shortcomings - and also different things they do well.

The workflow offered by TI and TINA-TI is more plug-n-play than that offered if I wanted to import TI's models into LTspice, so I use TINA-TI. Same for LT parts and LTspice.

Probably by me: but I use frypan to preheat PCB to 140C and then hot air gun from above to actually melt the paste to liquid metal.

Good to know. Thanks.

Just keep in mind that the temperature profile specified in the data sheet is there to prevent the parts from cracking during the soldering process. The assembly place I use uses a reflow oven with 13 (thirteen) temperature zones and Xray the solder joints after to verify their work. Many of the Far Eastern outfits use 3-zone ovens ... and get lower yield as result.

One of the guys I used to work with got quite good at soldering LLP/QFN-48s using a hot plate and a heat gun. He hit about 80% yield. Granted, that was on clock-jitter cleaners with 160 dBc noise floors and such, so perhaps more demanding than a headphone amp. The "low" yield and uncertainty around the process was enough to nudge us to eventually blow a bit over $20k on a METCAL reflow system. :)

Tom
 
Founder of XSA-Labs
Joined 2012
Paid Member
For prototypes, I have been lucky enough not to crack any ICs from too much heat. I use an IR thermometer to monitor temps on the board and try to not exceed the profile limits and at the same time to keep it hot ong enough within the soldet’s particular melt temperature soak (usually 10 seconds). The one tool that has improved yield and ease of use is the i-Extruder digitally controlled paste applicator. It’s great for neat joints and let’s you work complex boards section by section to allow testing before adding more parts.
 
I would strongly suggest to balance or minimize impedances next to the first opamp inputs. There are two reasons to do this. The first one: since your amplifier is non-inverting, it will have a significant common mode signal present at its inputs. This will degradate performance of the circuitry compare to the case when those are balanced or small. Specta below show grows of distortions going from low source impedance of 25Ohm - the green traces, to 1.6k balanced - the yellow traces and the red traces - unbalanced 2.2k/1.6k like in you schematic.

19kHz+20kHz 1.41Vp:
HonmHSp.png


10kHz 1Vrms:
CAujpFc.png


1kHz 1Vrms:
ccTssWc.png


The second reason is that impedance imbalance severely degradates opamp PSRR, essentially at the same rate it increases common mode related distortions. If you did this on the first place, quite likely you would not need to fiddle with supply caps ESR for minimizing distortion. And yes, your beloved model knows nothing about this wonderful effect :)
 
Last edited:
Founder of XSA-Labs
Joined 2012
Paid Member
Are you suggesting converting the amplifier into a balanced input drive? Most people don’t have balanced drive sources and will be connecting this amp to a single ended RCA input. Having balanced drive will require a 3pin XLR (or similar input) and require grounding of negative pin when not in use. The input impedance is 100Mohm//7pF and CM impedance is 6 Gohm//1.5pF, so I assume this means it is a FET stage and very high impedance. Pushing current through a 2k2 should be plenty of drive for this input stage. I don’t think lower impedance is necessary. Recall this is a consumer mass market headphone amp chip designed to be used in portable DAPs and cell phones. They are not going to require low impedance drive in those use-cases.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.