DIY 4.2 amplifier with DSP
The motivation behind this project:
I bought a Cambridge Soundworks FPS1000 system off a 2nd hand website. I always wanted this system since I got a 2.1 version 6 years ago. it produced an impressive amount of output despite it's power consumption and size.
Although the system worked well, there were some problems:
- many capacitors were leaking, requiring replacement
- the bass was really lacking, in terms of bass extension, and the subwoofer enclosure was tuned poorly, resulting in port noise
I addressed the first problem by recapping the whole board but the 2nd problem was fixed by using an external subwoofer and adjusting the high pass filter on the board. I also changed the op amps (I used two JRC2043 op amps for the 4 satellite channels and an RC4559 for the subwoofer in the end, not the NE5532s that I showed above, the originals were M5218s). The system played extremely well - the satellite speakers produced extremely high SPL when needed (105 decibels near the field using 4 2-inch transducers is an impressive feat given how the amplifier is only rated to give 3.5W per satellite channel), and the bass sounded extremely powerful with authority.
Images of recapped board
3D printed enclosure I made for the board, so that I can use it with a external subwoofer:
Despite this, I felt that a couple of things can be improved:
- Due to the use of transistors (in particular, H945 NPN transistors for preamplification and high-pass filtering), I heard a lot of crossover distortion and phasing issues as transistors are not the best technology compared to op-amps for such active crossovers
- although the satellite speakers sounded loud and clear, at high volumes the subwoofer channel had some very aggressive limiting, reducing the power extremely quick when the amplifier started to clip by even a bit. It took 3 seconds for the compressor to release and return to full volume. This issue is due to both poor compression circuitry and low power for the subwoofer channel
-there was little upper bass after changing the values for the high pass, resulting in a gap in the 80-150hz range.
How I addressed this by creating a board myself:
I decided to create an amplifier board using 2 STA540 amplifier ICs like last time, this time using digital preamplifier processing using a Wondom ADAU1701 DSP board. This allowed full customizability of sound.
Measuring and replicating the response of satallite speaker EQ:
I could tell from the beginning that the speakers are using EQ to improve the high-end response (above 5khz) due to the natural roll off of a full range driver, as well as a high Q high pass to give the satellites some degree of bass. By using an oscilloscope and excel, I created a plot of the response curve of the amp.
(I applied a log function to convert the voltages to decibels)
The EQ was simple, using 3 main filters:
High pass filter: frequency = 148hz, Q = 1.85
Peaking filter: frequency = 11000hz, Q = 0.3, Gain = 4.5dB
low pass filter: frequency = 19500hz, Q = 1.67
I used a BRU5 amplifier using ACPworkbench for this testing process
Sigmastudio project:
Overall capture
Input stage design. First off there is a DC blocker to block even the slightest of DC since it would be amplified by later EQ stages. This is fed into a compressor which acts as a noise gate:
After that, the signal is amplified by 22db (gain stage), allowing for wide input sources to be used as well as to maximize the volume seen later on - taking advantage of the headroom bits (allowing for 24db without internal clipping). This is placed after the noise gate so that even at high gain, there will be no hiss when nothing is playing and there will be no cutting in/out at low volume levels. After the gain and input stage, this is fed into a volume control block which uses the potentiometer readings to adjust the main volume.
Satellite channel processing - overall
First, a high pass filter is used shown below:
This initial HPF removes bass signals that would affect the output level of compressors coming later on. This goes into a crossover module and compressors shown below - forming a multiband compressor:
The crossover above splits the signal through a 48db/oct pair of HPF and LPF for multi-band compressor coming below:
Subwoofer channel processing - main view
First off, the signals are fed into a summer mixer, combining both L and R channels, and go through a volume control module so that subwoofer levels can be adjusted through a potentiometer. An 800hz low pass is used so that only low frequencies go beyond this stage. A compressor is used to reduce levels from the gain stage to a maximum of 0db, showcased below:
After the compressors, the signal is then routed through a parametric eq:
"Wow, that is a ton of boost" you may comment! This is 20 db of gain in this eq! Logitech does this with their subwoofers in their 2.1 systems - allowing for a very wide adjustment for their bass compensation knob. However, this is at maximum gain without any attenuation, so the gain would be much lower later on. there is a wide peaking filter, boosting frequencies from 40 to 70 significantly by 7db on top of 10 dB of gain. However, there is a 160hz high pass as that is the crossover frequency of the satellites. To remove the muddiness and boominess causing fatigue, I used a notch filter to attenuate frequencies around 130hz and beyond as those are what cause a cheap, resonant, fatiguing, and earache-inducing subwoofer satellite system. This is the EQ that thus provides the powerful and nice-sounding bass heard. After the bass processing, it is then fed into one more compressor to limit to prevent clipping, with the same settings as the latter photo, and 2 volume adjustment sliders to mute an unused sub channel to save power with such class A/B amplifiers. Yes, even unused channels can cause heat even with no load connected. After this, the processed audio is then outputted.
Final output processing shown above in photo
Amplifier design:
Of the 2 chips, one of the STA540 ICs was designated for the 4 satellite speakers, providing an output of 11w to each 4ohm channel at 10% THD. The other ICs were designated to the subwoofers - with 2 channels providing 38W at 4 ohm load each. Despite that, I am only going to use one of the sub-channels and the other will be muted in DSP to conserve power - but I may build a subwoofer utilizing both channels later on.
This time, I wanted to have a rather mature system for the amplifier, in particular with having popless startup and shutdown behavior (that means no loud turn-on or off thuds during power-on or random power loss situations). To achieve this, I changed the SVR capacitor from 47uf recommended in the datasheet to 330uf to achieve a 3-second delay so that the amplifier will always turn on before the DSP no matter what. In addition, I used transistors (s9014) for controlling the standby pin to allow for a standby switch and for the DSP to control the amplifier status. This means that the DSP will only pull the pin high after it has been initialized for a certain delay, and pull it down when power is removed, preventing on-off transients even in the worst-case scenario. The transistor portion was inspired by a Logitech Z523 schematic, with a tina ti simulation I did shown below:
I also modified the sure DSP board, by replacing the input resistors from the 18k ones that were stock with the board to 10k ones, allowing for full-scale input at 1V RMS. It is important to take advantage of the full dynamic range of the DAC by allowing almost every device to saturate the inputs, as only pro audio devices can drive to the 2V level.
A screenshot of the datasheet showing how resistors affect maximum input voltage
Picture of the modified DSP board, with resistors R8 and R7 changed from 18k to 10k
Schematic:
Unlike the last time I made a design, I made sure to use proper gain staging. I thought that many pc speaker makers feed an excessively high signal input voltage into the amplifier to achieve an illusion of loudness due to clipping. However, I discovered by probing the original amplifier that it is only feeding a 0.5V signal into the amplifier- a value large enough to drive the amplifier to full power. The distortion and loudness happen in the amplification stage and not through the amplifier.
In addition, DSP units have only a certain signal-to-noise ratio and will produce excessive hiss if not used correctly. I used trial and error and ordered many resistors between 1k and 5k to find a value that allows the amplifier to be driven to full output at the 0db level of the DSP, through the use of the amplifier boards from the previous post. I found that 2.9k is the best value, allowing for full output even at 18V.
Here are the PCB layouts. I paid close attention to routing this time to reduce noise by keeping signal and power traces away from each other. I did not incorporate separate grounding with the use of the DSP as the ADAU1701 calls for a shared, continuous ground plane for both analog and digital components! That's a rarity in mixed-signal designs for sure! One thing to note was that I replaced the 7805 linear regulator with a switching drop-in replacement with a 500khz switching frequency to prevent switching noise from entering the audio
Final product:
Some changes I made:
Add reconstruction filter capacitor (1.5nf) on all 4 channel
Gain stage resistor: 2.7k
Remove R6, R5, R4, R3
SVR capasitor is 330uf
Individual IC filter cap now 1000uf
DSP filter cap now 470uf
Add fan regulator (12v) under the DSP module
change DSP 5V regulator to switch one
One important thing I learned was the importance of a reconstruction filter. In delta-sigma DACs like the adau1701 used, it is used to low pass and remove the high frequencies used to create the sound we hear. One thing that I did not realize until later was how the slight hiss I heard from the speakers connected to this amp were caused by a lack of reconstruction filters. I thought the hiss were caused by power supplies because when I used a separate Apple power brick to power the DSP, the hiss became less. I switched to a regulator with a higher frequency and although the hiss decreased, it was still very audible. I realized that the wondom board lacked a reconstruction filter, and tried adding one by placing a 1.5nf capacitor after the 2.7k gain stage resistor, forming a RC low pass of frequency 40khz. This completely removed the hiss, which could only be heard when placing your ear right next to the speaker, making the system completely silent.
Outcome:
Really, I am left speechless. It is really one of the best systems I heard so far since the beginning of my passion for audio. It is loud, powerful, and silent in terms of not having any hiss. It is really all I can ask for. I am so happy about this system and absolutely love listening to it.
Comments
Post a Comment