The Class D audio amplifier - Basics (1/3)
HTML-код
- Опубликовано: 11 июл 2024
- #164 In this video I continue looking at the various amplifier classes, by starting working on the switching amplifiers. First up - the class D audio amplifier. I look at why a switching amplifier would be better than a linear one, and then I start looking at how the class D amplifier turns linear signals into square waves. So I mainly look at how you can get binary and ternary PWM modulation.
Related videos:
Class A amplifier part 1 - • The Class A amplifier ...
Class A amplifier part 2 - • The Class A amplifier ...
Class B amplifier part 1 - • The Class B amplifier ...
Class B amplifier part 2a - • The Class B amplifier ...
Class B amplifier part 2b - • The Class B amplifier ...
Class C amplifier part 1 - • The RF Class C amplifi...
Class C amplifier part 2 - • The RF Class C amplifi...
Class D audio amplifier part 1 - • The Class D audio ampl...
Class D audio amplifier part 2 - • The Class D audio ampl...
Class D audio amplifier part 3 - • The Class D audio ampl...
Class D RF amplifier part 1 - • The Class D RF amplif...
Class D RF amplifier part 2 - • The Class D RF amplifi...
Further reading:
www.ti.com.cn/cn/lit/an/sloa1...
www.infineon.com/dgdl/an-1071...
Special Thanks to all my supporters on Patreon! Especially Ralf B., Paul Pr. Richard, Jonathan Alvarado and Tosta!
If you liked this video be sure to check out my other videos and you can also subscribe to be up to date with all the new ones!
If you want to support the creation of more and better videos please consider checking out: / feszelectronics 
Наука
Idk why videos with people doing dumb stuff are more popular than channels like this where you actually learn something
Normal distribution of population 🙂
To be fair, class D amplifiers and analogue electronics in general are a pretty niche thing. Even well-read scholars aren't usually watching videos about very technical subjects.
Excellent videos. Thanks for making these well tailored content delivered at a perfect pace. I watched all three videos in the series in one go. The content is delivered with such great clarity. I got so many concepts refreshed and also many new concepts learned.
Been stuck on a broken class D amp, thank god i found this series tho, amazing video!
Fascinating! These are the amplifiers I've been waiting for! 👌
Are you an accountant?
Class-D amplifiers - unmatched efficiency and power delivery. However without quality output filter sometimes unbearable to listen to. Cant wait for the part 2 when he explains filtering !
@@BOT_0x76DE45AB No. Electronic Engineer. But doing CPU design, high speed digital, FPGA, etc., and not audio amps! 😂
Simply excellent! Thank you.
Love your vids
very informative...
thanks.
Thanks for this really interesting video looking at this basics again, someone like me has been dwelling on this topic for a while but more on self-oscillating topology (you showing a clocked modulator). For me one of the hardest subject that forms part of this journey is the control system feedback topology, modern efficient class-d will use self-oscillating control loops: which will require the following:
- a double pole integrator.
- a comparator with low propagation delay. (why? each component added influences the positive feedback phase-shift which will increase/decrease switching frequency)
- stability analysis across the whole audio spectrum under load.
- dynamic system analysis under load (notably near DC).
- poles/zeros needs to be place for correct reliability. (under different dynamic Z loads 2, 4, 8 ohms).
- fully differential input stage so common mode noise isn't amplified by the power stage. (again poles at the input stage for too high and too low frequency this applies to the main feedback loop as well).
- simple DC offset removal using x2 IN4148 diodes that clips the input when anything at DC is detected.
- the math is hard, many loops with different stability targets!
I would love to see this incredibly interesting video progress, few on RUclips covered above subject matter.
Hi, could you please put some links to those videos that explain those subjects? Thank you
@@jorpese1 Hi, the topics are sparse and found fragmented in papers on the internet I would recommend reading patents, university papers and lastly this guy: Bruno Putzeys he's work in the class d space is enormous study he's papers and UcD amplifier from there you will find and deduce other paths of knowledge, control theory is a separate subject you can find many many areas of study, the trick is how they overlap in class-d there is no one source answer.
@@yaghiyahbrenner8902 I'll check Bruno's work.Thank you very much!
先赞再看! love from China ❤
@aaaaa and then being able to do better reverse engineering. just kidding.
I've tested a so called filterless class d amplifier diy module (with Yamaha YDA138) and practically wiped out the reception of my favourite local FM station: West City Radio at 88.8MHz. Sorry if I did a spoiler for your future videos regarding this amplifier class.
Well, the filter is not needed for functionality... De aia se prinde asa rau postul prin oras? :D
Very interesting. To me this concept could open up a theory I've been struggling to do. If you use say PWM to amplify a guitar signal - Can you use a unit for example a ESP32 to change the incoming PWM signal and multiply the pulses by any number 2 to say 5 and with all your filtering on the output end, get a pitch shift up or down with still good quality??
Hi @Fesz Electronics: In AD modulation (min 11:04 video) when i simulate current through the Load2 the signal is second order system (damping) and not an oscillating system like the video. Why ?. Is there any initial condition in the Edit Simulation Command ?
I was performing a 10ms transient simulation; other initial conditions where I think to start measurement after 5ms. If there is switching before the filter - then even if the filter has good attenuation and it is damped, some of the oscillation will still pass trough. For ex: if the filter has -60dB attenuation at 100KHz, then a 100KHz 1V signal being injected will still pass trough at a value of 1mV.
What about a signal of varying amplitude, wont there be data loss for quite signals
Do u have real schematic for class D amp?
I really want to know about mosfet u use? & should we use mosfet gate driver IC?
Cz when i use mosfet for my final amp, the amplitude of square wave output is = to comparator output.
FesZ, is the DC offset on the input sine to demonstrate DC bias does not effect the driver?
The input stage opamps require that the signal be centered around half supply voltage. When I started working on the design, I tried to add this voltage offset into the input signal directly, but then went for the capacitor and input resistor divider; I think I forgot to remove the offset in the signal... Anyway, for the simulation at least, having the offset in the signal source helps get the amplifier into a stable state faster since it does not require time to charge the first capacitor.
Hi, great video. I have seen your video regarding class C amplifier. In class C amplifier the one feature is that the output voltage is twice the max. supply voltage. This feature fits my use case, since i am looking for an ampifier with voltage boosting. Is there any chance to get use of this feature also with class D by using an LC tank ? I have the option to give directly from the µC two inverted controll signals - so a class D is useful. Or a combination class C and D ?
You can use an H bridge output power stage - two stages each driving one side of the load in phase opposition; in some applications its also called a BTL (bridge tied load); this can work on both D or AB/B, and it will expose the load to twice the supply voltage but you need to check if your application can accept that none of the pins of the load are connected to GND.
@@FesZElectronics Thanks a lot ! You saved my sleepless nigths sitting on LTSpice. It works fine, at least with ideal components :) ;) By the way i faced a problem. But i am not sure if this is a really problem. On the single-ended sides of the differential load i measure my output signal, twice in amplitude, correctly filtered out. But on the single-ended signals i see that there is an DC offset. So the signal is swinging between -xV and +xV. The differential signal is important for the load. But is this behaviour normal for the single-ended signals ? How I can improve that ? My goal is to have an swing on the single-ended sigde always on the positive axis, because i want to track also the single-ended signlas with the ADC. It would great when you can give some hints or tricks. Thanks in advance :)
Like your videos, but would sometimes love some more basic explanations.
For instance when using LTSpice, since my Mac version only has a very basic interface I sometimes struggle to grasp how you set up a simulation and how to read the results.
I am almost certain that LT Spice on Mac has the same functionality as on Windows. At least the release version number is almost the same. So perhaps the "basic interface" you're seeing is because you haven't discovered some aspect of how to use the user interface, which is a bit non-standard. But there are LT Spice tutorials on RUclips, including previously on this channel.
Sending support idol
Please make videos on LTspice simulation of closed loop buck and boost converter
Reminds me of VFD motor controllers
A loudspeaker is basically a linear motor atached to a cone
can you share the symbols used in the BD schematic? you've got some un-labeled op-amps and gate/sw drivers.
I have watched three or four different versions on RUclips explaining this everybody is good at telling about the pulse width modulation but I don’t see anybody explaining how the amplitude information is delivered… If these devices only have two states on and off how does it know 100 millie volt input from a one volt input I guess I will have to set one up and look at it myself with a scope
What aspect of the digital signal is changing to reflect the change in input level?
You are right, there are only 2 states, "on" and "off", but these change at a very high frequency (say 400KHz); the analog information is always at a much lower frequency (say 20KHz); using a filter, you average the high frequency signal to obtain the low frequency one; at the output of the amplifier you never see the "on/off" states, only the long term average which is proportional to the duty cycle variation.
I have done a 5 level with 4 half bridges, however, I do not now how to calculate the lowpass filer because the inductors are summing on the output side into the capacitor who belong to that low pass, I see it that the 4 bridges who are low im pedance do let the inductors as paralell, what means I need 4 times the calculation outcome of a normal low pass, I have best results with bessel, I get -110dB 1khz 20 volts out, and -90dB 20 khz 20 volts out, I have design a feedforward system with two PI correctors. I have 90dB openloop 1 Khz. and 40dB on 20 khz. Have all done in simulator, I am a bad calculator.
Class-D is okay in terms of Power Efficiency, not good in terms of Audio Quality, Poor damping factor particularly at high's. Not a fan though.
But FesZ your videos are really good man. fan of your work
Is there any option / idea to class D amp sounds better
Can a single voltage source be used in full bridge class d amplifier?
Yes, that is one of its advantages.
Half bridge half of the voltage hhh, so if we want to amplify an ac generated by a dac we can use this technic first quantized then amplify then filter out. But part of driving the mosfet still unclear 🤔
FesZ, in the waveform for the full bride single power supply, your waveform is swing from +10 to -10 with only a +10 supply, how/
Its part of the reason for using a full bridge - when left side is pulling the load high, and the right side is pulling it low, the load sees +10V; when the bridge switches, and the left side side is pulling low, and the right side is high, the load sees -10V; "-" here is relative because its in reference to the "+" seen before; current is going the other way trough the load while its still exposed to the same absolute 10V. Putting the two halves together you get the full swing of +/-10V.
no Self-oscilating?
3:30 is this really so? The fourier transform of the square wave are harmonics of a wide frequency band. Low-pass filter cuts high frequency harmonics, leaving only low frequencies. That is, the output of the LPF is a sum of low frequency harmonics that do not make up a DC signal. Correct me if I'm wrong.
It is true that the decomposition of a square wave is a series of sine waves, however, there is also a DC component in there - if the initial square wave is not centered around zero (like +1v to -1v) but rather goes from 0 to 1 V, the decomposition will be a set of sine waves centered around 0 plus a DC bit which gives the offset. The LPF has a corner frequency below the fundamental of the square wave, so only this DC offset passes trough.
@@FesZElectronics I tried simulating in LTspice a square wave passing through LPH. Unfourtunately I can't see any DC offset, only "smoothed" square wave. Can you please point out what I am doing incorrectly? The netlist of the simulation is below.
V1 P001 0 PULSE(0 2 0 0.01u 0.01u 5u 10u)
R1 P002 P001 10
C1 P002 0 100n
.tran 20u
.backanno
.end
In this case the square wave fundamental is at 100KHz and the filter corner (-3dB point) is at 159KHz; there is no attenuation at the square wave fundamental; if the corner however is far lower (ex: R=10k; C=100nF fc=159Hz) then the filter output will be a DC at 1V, after a small stabilization time. That sort of filter will extract the "DC" from the square wave.
@@FesZElectronics oh, now I see it, thanks! In both input(before LPF) and output(after LPF) signals there is a fundamental(0 frequency) harmonic. If corner frequency is really close to the fundamental "DC" harmonic, than it becomes more prominent. However, I don't get how it works from the "mathimatical" equations perspective. Can you please give me a hint on it? As far as I know, LPF is an intergrating elements, or, in other words, the formula that describes LPF is the following: Uin = IR + 1/C * integral(I*dt). As I = C * dUc/dt, than this equation can be rewriteng in the following form: Uin = RC*dUc/dt + Uc. I tried solving this differential equation with bernoulli method so I can obtain an output signal that looks like sigmoid centered around 0 in xy-plane and which is "modulatet" by a high-frequency triangle(rising and falling slope, as a slope is a result of integration a constant, or Von, voltage), but it didn't work out.
Salut,
De ce ai pus follow-erii U3 și U5?
FesZ, I've got the model inputted, but when I run the simulation, I have not been able to get a solution. Do your run take a long time to finish. I have tried using the .ic option, but does not have. I've tried finding information on convergence to a solution.
Which exact model are you using? if you are referring to the op-amp maybe you can change it just to see if there is an issue with that particular model.
@@FesZElectronics I got the half bridge to work and the first full bridge to work, but when I add the 3 LT1721 op amps to get your model that has the second 10K divider doesn't reach a solution. It is the model at video time 13:10 or 12:30. Thanks
Ah, now I understand. Think I had the same issue as you are having, so the 3 extra opamps - U3, U4, U5 are not the LT1721 but rather the "UniversalOpamp" - this is some generic model in the Ltspice library, right at the end, and if you right click it, on the "Value2" line, I changed it to "Avol=100Meg GBW=100Meg Vos=0"; not sure this is need though.
@@FesZElectronics thanks, I'll give that a try. I thought that might be the case, but the numbering lead me down the wrong path. Thank you for our help.
FesZ, sorry, operator Head space, thanks
Great video but only able to understand for 6 minutes
The picture does not represent a real situation. Let’s say we have a 20bit resolution. That means about 1 mln different levels from peak to peak. If we have a 800kHz oscillator, a width of one pulse must be as precise as 1 to 1 mln. That means the shortest impulse as 800GHz. Did you see such mosfets? So, in a reality that precision is achieved with a “chaotic” jumping around a desired level with wider and shorter impulses, controlled with a feedback.
And of course, at 20kHz, when we have only 40 impulses per a wave cycle, that’s physically impossible to achieve 20bits resolution. That’s why D-class is terrible sounding at high freqs.
True digital amplifiers can solve this problem. Because the impulse width can have only 32-256 different values. This is about 25-204 MHz. An absolutely real number to have a no-feedback channel without a jiggling signal.
@@nosferratu Yes, "width of the pulse is modulated continuously" but that modulation is still in a time domain because we still operate with 1 or -1 (of a different width). Check out a theory of a delta sigma. A D class amp with a feedback implements a D/S modulator with all that math behind. A PSU level only affect those 1/-1 making them 1.1/-0.9 for example. A rest of D/S remains the same. LC filter of a D amp itself has a cutoff frequency about 28-31kHz (some D amp schematics for an active speaker have only an EMI filter for 1-2MHz and use a loudspeaker coil inductor as an L, but that's not our case), and a loudness level cannot explain a bad high frequencies quality. That's about a low resolution of a modulator in a time domain.