[Suspended] 6 channel, high power Class D amplifier with advanced DSP capabilities

So I am currently working on a amplifier project. The motivation behind this project is to get a low noise, powerful DSP functionality, and high output to drive my personal speakers. I bought a Wondom JAB-5 board and I was really disappointed due to the loud hiss of the outputs, and how my complex DSP code often causes the ADAU1701 to run out of instructions (dynamic EQ, superbass, etc.)

The speakers that I am trying to drive with the amp:

As you may know if you read my previous articles, I really like professional grade speakers. I like the form factor, industrial design, sensitivity, and the sound signature itself of such speakers over home audio equipment. PA/ installed sound speakers have the transient response, slight brightness, and liveliness that I find most home audio speakers fail to deliver in my opinion.

Left: example of low end, home audio bookshelf speakers. Right: my JBL Professional Control 25-1 speakers with the speaker mesh removed. Notice the horn loaded tweeter that is used to control the directivity of the driver and increase the efficiency in terms of air coupling, compared to a dome tweeter. Although the cones look similar, the PA speaker has one of a lower moving mass, but sacrifices bass response for efficiency. 

I also appreciate the concept of headroom. Even though I may not listen to music at 110dB every time I play music (my neighbors will more than hate me), it is good to have the option to do so when appropriate). Since PA speakers are rugged and designed to be abused/carried around, I can take these speaker with me like a live DJ and play music at large rooms for everyone's enjoyment. Or I could use them in a outdoor space when needed. 

For those people that don't know speakers too well, think of it as owning a Truck (I use this analogy even though I am a sports car person over a truck individual). You may not always use the towing/bed capacity of a large pickup truck but if you have something massive to carry (e.g. a large TV, furniture, moving from one place to another), you have the ability to do so easily. This is basically what I mean by why I like headroom since I like the ability to push out high volumes when I need to.

In addition, the ability to have greater headroom reduces distortion at lower volumes (~90db range). Most speakers will increase in distortion because of a greater cone excursion, but since PA speakers use light cones that can be easily driven by the voice coil inside the driver, it can get louder without too much cone movement and power, reducing THD.

- 4 satellite channels: JBL Professional Control 25-1 speakers: 8Ohm, 75W continuous power, 90dB sensitivity. I will arrange them as front left and right and have the rear pair as rear left and right. This is basically the same as a quadrophonic arrangement. When using a 2 channel source, I can either do a stereo 2x arrangement, or use ADI's upscaling technology from the DSP.

Fun fact: I got 5 of these speakers for 30$ on ebay, but paid 60$ for shipping. Buying these MSRP costs 500$ and the market price for these is around 160$ per pair. But getting 5 for less than 100$ in total is a steal, given how good they can sound with proper DSP and how loud they can play (108db continuous). Very impressive for a 5.25'' LF driver and a 0.75'' HF tweeter.

These are my favorite speakers and I can't think of anything that can produce mids/high that I like better than these! Love the form factor, sensitivity, and sound profile of these speakers!

- Subwoofer: QSC AD-S28TW

(I choose this speaker since even though it was expensive (200$ with quite expensive shipping), It was smaller than the JBL ASB6112 I wanted to get at first. The added benefit is that since there are 2x 8 inch drivers wired up in series as a 8 ohm speaker overall, I can modify the wiring so that 2 amp channels can individually drive the 2 drivers inside as a 4 ohm load. This means that the subwoofer would receive around 320W total, 160W per channel.)

A view of the 2 8'' speaker driver inside the subwoofer. It is interesting how the speaker drivers are arranged in order to conserve space.

 A picture of the solidering job I did to wire out both drivers as individual 4 ohm loads. This maximizes power transfer to the subwoofer. Weight is also reduced from 19.5kg as the heavy transformer and crossover is removed from the enclosure (why am I going to use this with a 70/100V application for home use and why use passive crossovers when you have DSP?) I think it weighs around 16kg now, close to the weight of the ASB6112 I wanted to get but accepts more power since it is a 2x 4Ohm load and not a single 8Ohm one.

I love this subwoofer and just like the JBL controls, I have never heard bass this fast ever in my life. Even if it doesn't play super low, I rather have fast bass than slow low bass. And what is the point of annoying my neighbors in my dorm when you can't really hear 22 or 30hz too well anyways? It would be more effective to use a missing fundamental effect (or ADI superbass) to reconstruct and trick your brain into hearing those ultra low notes, which I will do with the DSP to make this 8'' subwoofer sound like a 15'' in terms of perception of bass extension.

The shortcomings faced with the JAB-5 board:

- Poor SNR: I can hear hiss when nothing is playing. It seriously drives me crazy since the amplifier i built previously with a ADAU1701 chip is dead silent when nothing is playing, even though the 2 amps share the same DSP IC. I am sensitive to speaker hiss and it's something that bothers me about active speakers or poorly implemented class D amplifiers. This is even more annoying since one of the 4 channels has a louder and more annoying hiss than the other 3, in the 1khz range, which is in the realm where our ears are most sensitive! More annoying than a mosquito!

- Low instruction count: My dynamic DSP code takes up a good majority of the instructions (around 700 out of 1024), and this limits the amount of things that I can do more. I want to do stereo upmixing, different EQ profiles, and more but I am limited by the number of instructions per cycle.

- Lack of subwoofer channel: need external amp + DAC combo. Not optimal for my use case

How my new board will solve these issues:

Mitigating noise:

- Using a Class D amplifier IC with a high SNR: such as the TPA3255 or TPA3251. These chips have a signal to noise ratio, with the TPA3255 having a SNR of 112dB and the TPA3251 with 111dB of SNR, both in BTL mode. This is much higher than the TDA7498E chip, with a SNR of around 90 or so dB. The noise figure of the TDA7498E is similar to the TAS5630B, which is really bad.

- Using differential connections from the DAC to amplifier. This reduces the need to consider return currents/ground loops as the noise is canceled out between the positive and inverted (180 phase shift) outputs. Thus this cancels out noise that may have been carried and amplified by the TPA3255 if a SE connection was used & eliminate the chance of common mode noise or ground loops.

- Using chips with a high SNR:

- Paying attention to return currents:

- Using low ripple power supplies for components like the DAC, op-amps (if used),  and regulators related to amplifier

- Allow for digital inputs (such as USB, SPDIF) to eliminate noise by using a normal ADC.

- Use components with high PSSR to reject PSU switching noise

Lack of instructions:

- I will use a high performance DSP IC: The ADAU1466, which can handle way more instructions (at least 6x as more compared to the ADAU1701), and this allows me great flexibility in how I want to route my signal path

Other important considerations:

- No pop or click noise upon power on/ power down: and this includes unexpected power loss. The amplifier must not make any of these transients under normal circumstances. This is going to be accomplished by using a AND gate IC, which will pull the RESET pin high to turn the amp on ONLY if the power rails are stable (from a power supervisor IC). the DSP is initialized and ready to output sound, and the 555 timer finished doing its counting down for a 5 second delay as a last measure. 

- A volume control, bass volume, and 2 other potentiometers are the minimum potentiometers that are required for the analog version of this amplifier

- Amplifier must be very reliable and not have the chance of blowing up under high power, heavy use circumstances. This means routing thick traces and following TI's board routing PCB requirements to ensure that traces can handle the high current needed for this power amplifier

Potential chips to be used: The ones with the strikeout formatting means that the chips are completely out of consideration and will not be used in the design for reasons described next to it.

Amplifier chips:

- TI TPA3255 - verified by me to have the noise requirements to keep me happy. Excellent sound!

- TI TPA3251 - Basically the TPA3255 little brother. Both share the same pinout, but runs at a lower voltage (36V vs 51V), but with a lower RDSon, leading to lower heat dissipation at 36V, which is what I will be running my amps at. But the TPA3255 will be used if it is cheaper than the TPA3251 since derating will help with long term longevity of the amplifier.

- TI TAS3251 - fully digital input stage (I2S), allowing me to eliminate the DAC IC. Digital version of the TPA3251. However it is I2C controlled, increasing complexity, and requires a special PON and PDN sequence (MUTE for both output stage and internal DAC).

- TI TPA3223 - Less considered because of reports of amplifier blowing up (example 1, example 2), likeliness of pop issues even if reset is used for PON and PDN sequences (link)

- TAS6484 - 2.1Mhz switching! 4 channel output! 45V input! Fully digital! But requires a NDA to get complete datasheets for. As this is supposed to be a open-source project after I build it so that you don't need to go through the struggle I went into to build a high power, DSP, multichannel amp, this is out of the question even though this may have been the best candidate. It also requires MCU control I think, which throws it out of consideration, just like with the TAS3251.

DAC chips:

- TI PCM5242 (eliminate op-amps, very low noise, however i2s would be a challenge due to the slim chance that I2C is needed to communicate with the chips if the DSP is unable to output a I2S stream that follows the strict timing requirements required by this DAC). Update: the PCM5242 is just like a PCM5122, but differential. In hardware mode, the PCM5242 should behave like a PCM5102.

- TI PCM5102A (easier to use than PCM5242, however requires a digital / analog ground plane. Not a differential output, so requires one inverting op-amp per output channel). Increases grounding complexity and component count.

- ADI AD1939 (integrated ADC, integrated DAC, rather high SNR, single IC. However, requires 3 op amps per channel (1 to convert diff -> SE output from DAC, 1 to amplify the signal to 2v RMS, 1 in a inverting configuration to drive the inverting side of the amp). Op-amps increases the chance of pops/clicks when starting up the amplifier, especially in a virtual ground configuration. Cost is a plus on this chip, but I don't trust ADI DACs because of what happened to my ADAU1701 with more hiss on one channel. This IC is just trouble.

- AKM4458 (excellent SNR for both ADC and DAC, cheap and well integrated, but requires external op-amps, resulting in more components and a larger PCB size. Complicated.)

ADC chips: 

- TI PCM1802: Decent enough SNR for a ADC especially with a noise gate, simple design.

- TI PCM1865: Very good SNR, but requires external MCU. Adds complexity to design.

USB input IC:

- TI PCM2706: may be enough for use case, but need to pay attention to potential of USB noise on audio signals. Only 16bit audio and is only 2 channel. May need 2 with a USB splitter?

- XMOS USB audio chip - great, but very expensive. I don't need 192K sound since my DSP is running at 48k, which is good enough for me. I highly believe in the nyquist theorem, and that science is the way to go about sound, but I do understand the need for higher sample rates in a recording environment. I rather have more DSP instructions than that, since I am just playing back music, 99.9% of the time from a digital source that is 44.1K or 48K anyways.

DSP Chips:

- ADAU1467 (optimal)

- ADAU1466 (like the ADAU1467, but with less IO)

- ADAU1452 (failback if cannot source or design with above chip)

Op-Amps (if used):

- TI NE5532: favorite op amp for me: low noise, great sound in my opinion, cost effective

- TI OPA1632: Good noise performance, outstanding for musical playback

Microcontroller (if used):

- ST STM32 series (Maybe a STM32F407?) This adds too much complexity in the first version, will be a external board that communicates with the DSP amp via I2C. Will stick to analog control knobs for now.

Amplifier block diagram:

So this is how I am planning for things to work. So first off, there are multiple inputs that can be switched by using a physical switch. This either pulls different GPIO pins on the DSP core itself or does this to the MCU and tells it via I2C to switch inputs via a slew mux (to prevent pops or clicks when doing such operation). Or it does this: Mute for 0.5s -> switch input -> wait for 0.5s and then unmute input to prevent annoying pops or clicks.

The DSP core does different types of processing (e.g surround sound mode, EQ, dynamic bass boost, missing fundamental generation, filters, EQ, volume control, and more) to process the sound in a way that I or the end user sound desirable. Everything can be configured in sigma-studio and changed using a computer without the need to switch physical components, as with analog audio circuitry (op-amps)

The DSP then outputs the processed signals to the DACs. Differential DACs (PCM5242) will output a differential signal to the different amplifiers.

The Amplifiers (TPA325x) are class D and there are 3 of these chips on the PCB. 2 of the 3 chips will be configured as BTL (bridged tied load), and could output around 70W of power at 8Ohms and 36 volts. The Subwoofer amplifier is configured as BTL and outputs 140W per driver or around 280W for both drivers. All channels are still identical, but the subwoofer channels have larger capacitors (2200uF vs 1000uF) to better support bass transients. 

On the output side of the amp ICs, a LC filters, followed by some RF/EMI filters will be used. After this, a set of catch diodes (connected to VCC and another connected to GND) will be used to prevent the amplifier from being damaged under no-load conditions, something that Wondom and some other brands do on their class D amplifier to protect their amps. Although I will give the option to turn off the amplifiers with the oscillators configured as slave (rear surround and subwoofer amp), it is better to be safe than sorry.

The RESET pin on the IC will be used as a standby pin, similar to how I used it on the STmicro STA540, to prevent pops and clicks and to allow the chip to enter a low power state. A power supervisor will control this pin, in addition to the DSP and 555 timer to eliminate this issue. Also, the Xsmt pin on the DACs will be used to further prevent the manifestation of this problem by prevent pops from forming from the signal source.

Update 6/29/2025:

Well it is unfortunate that I haven't updated this article since 5/8/2025. In around these 2 months, I made major progress with the PCB. The final IC selection is as followed:

Amp IC: TPA3255 (3x)

DAC IC: PCM5242

DSP: ADAU1466 module from Midiworx

ADC: PCM1802

USB to I2S: PCM2706

Reasons for why one could use a TPA3251 instead of a TPA3255, especially if the TPA3251 is cheaper:

So the reason I could go with the TPA3251 is because I don't need the output capability of the TPA3255. There is no way I can find a power supply that can output 1800+ watts at 51V. Since I am running the amp at 36V, the TPA3251 seems more suited since the RDSon is lower for the output mosfet (85 for the TPA3255 vs 60 for the TPA3251). This results in less heat produced.

In addition, the inductors I choose are the Sagami 7W14A. These are inductors that have a Irms of 7.6A and a Isat of 11.5A. I don't think these dual inductors will like anything near 51V at all, let alone 45V. So why think about choosing a amp IC that can fry the on-board components?

Also, even if I had inductors that could carry all 1800W of power total through all 6 amp chips, imagine how large the trace would be for the power. For even 36V, the power trace on the exposed layer would need to be 30mm to support 33.3A at a 10 degree rise, a large amount. So going higher in voltage will be impractical for my needs.

Also, if I had all of those factors sorted, this type of power will fry my speakers. I could destroy my subwoofer if I put in the rated 175W that this IC can output into a speaker rated for 125W continuous per driver (250W/2). So limiting will have to be done already even at 36V. For my control 25-1s, I would have the right power (88W), which has to be limited to 75W of continuous operation. So the amplifier is already powerful enough for my needs.

Therefore the TPA3251 will be sufficient, but for the sake of longevity, reduced cost, and a minuscule improvement in noise performance, I would still choose the TPA3255 for now. But if the TPA3251 is cheaper in your area, I suggest using that as a drop in replacement for the TPA3255 since it will save you money.

The reason for using a ADAU1466 board module instead of having it on the PCB itself:

So at first, I wanted to use a 2 layer board to reduce the costs of building the PCB. However the ADAU14xx chips require 4 layer PCBs for cooling and also signal integrity. So I decided to use a built ADAU1466 module to avoid this issue. But I am now using a 4 layer PCB and could've had the DSP built onto to PCB.

However, the project is already extremely complex, and designing a whole DSP module would be extremely difficult, if not risky since there is a good chance that my design will not work. So using a DSP module reduces the chance for error in a already extremely complex project.

Important design considerations made:

Mixed signal ICs

ADCs:

One important things for the ADCs is that the ground of the single ended signals shall be equal to the ground reference for the DACs, not the switching noise the class D amplifiers are making. To shield the sensitive return paths for the ADCs, I had to make a small analog ground region (or "AGND") for the analog signal path to keep that section of the ground clean. This results in a clear signal.

Of course, the Analog ground is tied to the rest of the ground plane and is not isolated from the rest of the circuit. 

DACs:

Surprisingly, the DACs are happy to be on the same ground plane as the TPA3251, since the output signal is differential. According to a digital input reference design amp by TI, the schematic shows the DAC sharing the same ground as the amp, and the common ground design is also shown in the EVM. This is further re-enforced in a design with multiple DACs (PCM5252 - a PCM5242 but with extra DSP), the TPA3244 Dolby Atmos sound bar reference design. Thus, it is safe to conclude that sharing the common ground will not introduce common mode noise, at it is simply canceled out, unlike in single ended designs that require a star ground to the analog pin.

Power:

For all mixed signal ICs, analog power is provided by a linear regulator, while digital power is provided by a switching regulator shared with the DSP. This is to ensure that analog stages receive quiet and low ripple power so that noise will not be injected into the signal path.

Amplifiers:

Power:

The 12V supply is derived from a linear regulator, to ensure quiet power. The rationale is the same as the reason for linear power supplies for the DACs/ADC's.

LC filters - inductors:

The Sagami 7W14A-100M (dual 10uh inductors) are used in this design since a small, but high power handling inductor is needed. Lsat is rated at 11.5A, while Irms is rated at 7.6A at a 40 degree rise. Although this seems hot, the amplifier uses active cooling with a massive heatsink (despite being a class D design) and fans will also cool down the inductors in addition to the amp chips, so temperature is reduced. Since I was unable to source these inductors from Mouser or Digikey, I turned to a compatible drop-in replacement from Aliexpress (1414 10uh), which datasheet claims to have same specs as the Sagami inductors. Note, the datasheet claims better specs than the listing, and I believe the spec sheet as I believe that they created this inductor as a alternative to the 7W14A for manufacturers at a lower cost. Engineers usually get very upset if the datasheets puts out wrong claims, but I would still take it with a grain of salt. However, I am not responsable for amplifier failure at all, and you make choices at your own risk.

No load protection:

Since LC filters with out a load will cause large amounts of back EMF/flyback effect that are much higher than the rail voltage (PVDD), this has the potential to damage both the output filtering stage and also the amp IC itself. Catch diodes are added to the output of the LC filter to clamp the maximum oscillations of the LC filter to the rail voltage. The catch diodes are positioned after the LC filter since placing it before it would be redundant since MOSFETS themselves have internal body diodes that serve the same purpose. SS16AF diodes were used for this application

Cooling:

*I am using 3x 50mm fans, with a 150mm wide * 50mm width * 50mm tall heatsink that was designed with a Class AB amplfier, to keep things silent yet cool, unlike my old Wondom amplifier.

Current schematics as of 8/6/2025:

Layout:

Layer 1 - Signals and AMP VCC:

Layer 2: GND

Layer 3 - Power plane:    

Layer 4: Signals

3D:

PCB front

PCB rear

TPA3255 closeup

Rear - output stage zobel network and no-load protection diodes

Closeup: PCM5242 DACs

Closeup: PCM1802 ADCs

Closeup: ADAU1466 3D model of a Midiworx DSP module I created

Mechanical design:

Here is the heatsink spacer for the AMP ics. The height is 3mm and there are going to be 2mm spacer in between the AMP chips and the heatsink. The spacers are secured using thermal glue to the heatsink while thermal paste is used between the spacers and the chips for easy removal.

Closeup of the chips and spacer

Closeup of mating between the 2 surfaces

Fan arrangement closeup: may replace 3x50mm fans with 3x40mm (40x20) Noctua fans for lower noise and due to space constraints. However, the large surface area should make up for the loss in fan CFM.