Multi-Zone Stereo Page
STM32 Design Contest Entry
This project will be published in Circuit Cellar Magazine Jan. / Feb. 2014 issues
The Schematics and PCB files are here
The STM32F embedded code is here

In 2010 I wanted to improve my coding skills, get into a new processor, and start a product development. A lot to take on all at once. So when the opportunity to enter the STM32 Challenge appeared in December '10, I decided to do it. Winter is a perfect time to do nerd projects. So what to design? I wanted to do something with both nerd and general appeal, something that I was familiar with and could use myself. Years ago I built a house-wide home stereo, capable of playing music anywhere in the house. I have wanted to update this system to newer technology. The original system sports:
It was built in the mid 90's, and powered by a Motorola 68hc11 processor and surplus graphics LCD driven by a custom FPGA. The CPU board had separate RAM, EPROM and I/O, and was hand-wired. The crosspoint was also hand-wired. The preamps were small home-etched single-sided PCBs, one per zone. The original system used a motley assortment of external amplifiers, collected over the years.
Here is the original system, now dusty and neglected.

Original

The new system should do all that plus:
Other features on the wish list:
3/26/11
I already had the basic code and LCD working when the boards arrived. I spent one long caffeine-and-pizza-powered weekend writing the control code for the crosspoints and preamps, front panel keypad. STM libraries are quite powerful, maybe too powerful. Fortunately there are good examples of how to use the basic and advanced features of each peripheral. For example, there are about 30 I2C functions, and 5-6 are required just for basic I2C master operation. If there is no I2C device present, then no acknowledge is sent and the example code just hangs waiting for acknowledgement. The debugger can then be stopped to find out where in the code it is hung. Since I have a bunch of I2C devices, not all may be connected at any time. having it hang trying to access a non-existent device is no fun. The best way is to poll the device, timeout if no acknowledge is present, and reconfigure variables to tell the control code it isn't there. I didn't have time to code all that so instead I made a constant for the number of preamps to control.

Most preamp functions are working including the keypad and IR remote control (not shown). The ethernet module is just sitting there. No hardware or software yet. There is still just one preamp channel loaded.

I prepared a video demonstration for the contest proof of concept stage and was selected for the finals, and the judging is in early May. Lots more board stuffing and coding to do.

4/15/11
Here is the first pass open breadboard. Across the top, L-R, are the crosspoint and the 2 quad preamp boards. Across the bottom: the CPU board with STM32 processor module and Wiznet Ethernet modules is just plugged in, not connected, the 128x64 LCD, and the 32 button keypad board.

system photo

Below is the Crosspoint board with 8 dual input buffer opamps and two 8x8 crosspoint chips. The crosspoints use 8 bits of control, provided by the IC on the bottom. It is an MCP32008 I2C GPIO port that drives the 8 parallel bits.

The 16 pin ribbon on the bottom supplies +/-12V and +5V power and I2C control from the CPU. The two wider ribbon cables supply the same power and I2C, plus 4 channels of stereo audio to each of the preamp boards. This board has ground planes on top and bottom to guard the audio signals from each other to reduce crosstalk.

Crosspoint Board

Below is a Quad Preamp board with just one channel populated. The preamp chips are ST TDA7439 which provide full function audio control. The red and grey film caps are for input coupling and tone control.  These preamp chips don't have I2C address selection, so logic on the left uses another MCP32008 and a CMOS switch to send I2C to just one of the 4 preamp chips. All audio signals are guarded from other channels and from the I2C control signals to reduce digital noise and channel-to-channel audio crosstalk.  The bottom PCB layer is almost solid ground plane. As a result, there is no audible I2C crosstalk.

Crosspoint board

4/20/2011
The project is a top ten finalists, and here is the prototype ready to ship to the STM32 Design Challenge. A front panel was built, and a power supply added. All 8 channels are working as are the IR remote control, the encoder knob, and the remote keypads.

Phase 3 system

5/8/2011
I'm happy to say the project won an honorable mention, one of three second prizes.  Meanwhile I have started building a second prototype.The Crosspoint and Preamp boards are stuffed and working. Most of the time was spent hand-wiring a smaller CPU board. The original CPU board (above) was incrementally built over several months, starting with the STM32 Discovery module and the LCD, slowly adding the I2C, Keypad, remote keypads, IR, encoder, etc. Hand wiring the second board has taken many hours, and I realized that I could have done it better, faster and smaller on an ExpressPCB Mini board for $80, for 3 bare boards.

Now on to building the final unit that will live in the living room. It needs audio power amplifiers and a real case. This version will have 4 stereo amplifiers and 4 line level outputs. Here is the block diagram.

Block Diagram

6/17/2011 LM3886 Amplifiers
The prototype is back from the ST design contest, and the next step is to build or buy amplifiers to drive speakers in the rooms of my house. I, like a lot of of the DIY audio community, am pleased with the LM3886 audio amplifier from National Semi. I have built up a few LM3886's over the years. My plan here is to build a modular multi-channel version to drive speakers around the house, that is cheap, simple and provides good performance. My goal idea is to build 4 channels of the amp on an ExpressPCB mini board. Here are my requirements for the amp:
One challenge with the LM3886 is that the package is designed so the PC board is at a right angle to the heat sink mounting plane. This type of mounting requires a fair amount of physical volume. A 12 channel amp would require a large case. To save space, I envisioned a design where the device was in the same plane as the board. I am happy with the results. The board mounts only with four screws in the mounting holes of the four LM38886's, which are sandwiched between the board and the heat sinks. The board can be flat mounted to virtually any flat heat sink with just four #4-40 screws. This requires re-bending the LM3886 leads, a simple operation with needle-nose pliers. I used the insulated package LM3886 so no insulators are required, just a bit of thermal compound on each device.  While searching Ebay for heat sinks. I found a nice Express-PCB-Mini-Board-sized one (3.75" x 2.75" x 1.25", search Ebay for "heat sink 3.75" ) and bought a few. It has a wide mounting flange on the bottom and even comes with two 6-32 tapped holes on the flange. It could not be more perfect.

Here is a 4 channel amp mounted to its heat sink.  The board uses a mix of surface mount parts (0805 and larger) and some thru-hole on the front side. There are only the LM3886's on the back. Film capacitors are used for  input coupling, and a Zobel network is used on the output. No output inductor is currently used, but there are provisions for one in case it is needed. For the amp DC power and speaker connections, pluggable .2" screw terminals are used  These are reliable, small, inexpensive, and easy to field connect.

4 Channel LM3886 Amplfier

The back side showing four LM3886's, ready for thermal compound and to be mounted to a heat sink.

Rear view

This is what 12 channels of amplification would look like. At this density a fan is needed. The speaker connectors will protrude from the rear panel (not shown).

Side View LM3886 Amp


Multi-Channel amplifiers require a ground trick
My goal is to build 8 or 12 channels in a small space and for short money. So it is desirable to use one bipolar DC power supply for all the amps. A problem is that most amp designs will crosstalk badly if you simply wire multiple amplifiers to a single power supply. The high output current from one amp channel induces audio voltage on its input ground pin. This causes the next amplifier's input ground to have the same voltage. So there is a voltage between one amplifier's input an the next. This voltage is multiplied by the amplifier gain (+26dB) and appears as crosstalk.  This crosstalk is a function of the resistance of the audio cable grounds and the length of the ground return wires between the power supply and the amps.  A spice simulation of this problem showed that sure enough, with normal wire sizes and PC traces, the crosstalk from one zone to the next is about -40dB, not very good. But by making an "input ground", separate from the speaker grounds, and tying the input grounds to the power grounds only at the power source, the crosstalk is more like -90dB. Electrically this requires a separate Ground return pin for each amplifier board, where the input, input circuitry and the feedback circuitry all return. Then these low current grounds are run back the the common power supply "star" ground. The result is that only very low currents flow through this input ground and the crosstalk is reduced to about -90-100 dB.  Here is the final board schematic showing IGND the separate input ground. Ideally there should be an IGND per amplifier or per zone, but a single connection per 4 amps will do the job well enough and simplifies the wiring.

Schematic

How much power?

For the main living room system, I have a separate high power amplifier and speakers. For the remote rooms,  the bedrooms, deck, basement, etc., typically these will be used at 5 watts per channel per room. Modern speakers are small and efficient. 30-40W amplifiers provide lots of headroom, preventing clipping. It is unlikely that all zones will be simultaneously driven to maximum power;  I have neighbors. My goal is to build 8 or 12 channels to drive 4 or 6 rooms. So at 5W per channel x 8 channels, that's only 40W of audio. That requires about 80W of DC power. If one or more rooms are turned up loud, they might use 20 or more watts each. And the transformer and power supply capacitors should have plenty of capacity for the occasional loud passage.  I suggest a 160VA transformer for 8 channels, 250VA for 12 channels. Audio purists may be concerned. Since the '70s I have designed and built many amplifiers and when designed 'by the book' to provide full power sine waves, the designs are far too conservative and heavily over-designed. The only time anything gets hot is when I am in in the test lab heating up 8 ohm load resistors with full power sine waves. In normal use, nothing even gets warm until you are about to go deaf. Real audio is different than sine waves. Sine waves have a peak to RMS ratio (crest factor) of 1.41:1. Real audio is closer to 5:1. So the system will have plenty of capacity for big bass notes but will stay cool.

But alas, there is a problem. I had planned to tightly pack these amplifiers, and to use the LM3886 mute function to control the power to each zone as needed. Then use a single small, quiet fan to cool them only when the temperature rises. Do you see my error yet?... The mistake is that I assumed the LM3886 mute function reduced the quiescent current of the amp. Nope, it just mutes the audio, in my opinion a fairly useless feature. Each amp draws about 50mA of current always, even when muted. At +/- 30V  each amp uses 60V x 50mA = 3W when it is powered. 8 amps draw 24W, 12 draw 36W. Pretty bad. With small heat sinks and close packing, the temperature rise is high enough to warrant a fan at all times. I am bothered by wasting 24 or 36W all the time. There is no easy way to turn off the DC to each amp. Doing so requires 2 high current transistors or a relay for each zone. The current boards use single power connections for 2 zones, so no easy way to control each zone.

My interim hack is to reduce the power transformer secondary voltage a bit (new transformer) and manually switch DC power to the rarely-use amps so they only draw current when they are being used. In my house, the most commonly used zones, the Living room, Den and 2 Bedrooms,  each have their own external amplifiers. The Kitchen and Basement are commonly used and so are always on. The other 2 zones (Deck and Bathroom) are rarely used and so I switch their power manually.   For now there will be a simple front panel switch to switch DC power to the amplifiers for those zones. This issue is motivating me to consider efficient, switching (Class-D) amplifiers. These will address the quiescent power issue, but I suspect will make the crosstalk issue much worse when operating from a single supply, since Class D amps draw lots of high frequency switching current.

Enclosure for the Preamp and Amps
With the preamp and 8 or 12 channels of amplifiers, I wanted the entire system to be mounted in a single enclosure. A Visio drawing below shows that this is possible in a 2U 19" rack , a 17" x 12" x 3.5" enclosure with 1" rack ears to make it 19" wide.  The advantages of a single case for both preamp and amps are:
The disadvantages:
Also if I ever add power amp DC power control by the preamp, having that wiring internally will be cleaner than doing it externally.  The trick to getting it all to fit is to stack the two Quad Preamp boards piggy-back style via spacers. The bottom one has 4 sets of RCA jacks to drive line level outputs to 4 zones. The top board has 4 x 4 pin headers for internal connection to the 4 stereo amplifiers. There is room for 12 channels (3 modules), but I expect to only use just 8 because that is what I need, and also to save power. The added height still fits nicely in a 3U (3.5" ) enclosure.

For the +/- 12V power to the preamps, small DC-DC converters will convert the +30V from the power amp supply to regulated +/-12V. This is not shown. 

For cooling, there will be exhaust holes in the rear panel near the speaker connections. Intake holes will be in the top cover, forward of the fan.

case
6/29/11 Getting there.
The main boards are mounted to the base and a new 19" front panel is in the works. Most of the wiring is in place and the system is working well!

Installed

Final

7/25/11 Installed, and Control code cleaned up.
The new front panel is installed and looks good. 8 channels of amplifiers (4 zones) are installed and wired up to speakers around the house. The new remote keypads are working. My family has attended the training class and they are enjoying music in the new zones as well as the old ones. 

There is no fan yet or DC power switching and so the amps get quite warm. The chassis consists only of a bottom and front panel.

The original quick-and-dirty control code was bothersome. The zone management was all manual. When you controlled one zone, it permanently stayed at that zone until you set it back. The expectation is that the front panel controls (IR, knob and front panel buttons)  are primarily for the living room zone. The original system had a default timer to reset the controls back to the main zone,  and this one needed one too.

The mute key caused an annoying problem. Having every key auto-repeat including zone select and source select was never a real problem since the audio output was barely disturbed. It simply selected the same zone or source over and over. But with the muting, you want to mute once and stay muted until the button is pressed again. It is the first function where auto-repeat is a problem. The old way of having zone and source select repeat as long as you held the control was inelegant. So now there is code to only repeat those functions that should be repeated: volume, balance and tone control. 

I addressed the control code. All of the controls now use a consistent data structure for de-bounce time and repeat time. It also has a time-out timer to reset the zone back to the default after 20 seconds. Keys no longer repeat by default, but only those keys that need to repeat do so. The user experience is much better, less twitchy now. It's amazing to me how much time, effort and code is needed just to make a handful of different buttons act nicely. I want them to all act similarly, and don't want to duplicate that code, so made a consistent set of code for the various controls. Arrays of control structures are your friends.

Another system issue is fixed: The various line level sources are pretty well matched, but they were higher level than expected. Also my main living room amp has more gain than the LM3886s (34dB vs. 26dB) and has very efficient speakers. So the lowest volume setting of -47dB on the TDA7439 preamp chips was too high for the living room. But the maximum level is just over clipping (and very loud!) so I am reluctant to just turn the gain down on that one zone. I suspected that -47dB of volume range was inadequate, and it is. Now the volume function uses some of the 'speaker level' range of the preamp chips to extend it to 60dB at settings below -47dB. There is also an input range adjust that can be used and has 2db steps. Using the speaker level for both balance and volume complicates the volume and balance setting code a bit, but 60dB is a decent control range. Maybe 70dB would be better? I also changed the volume steps from 1db to 2db so it takes less time (30 steps vs. 60) to adjust the volume. 2db steps work fine.

9/1/11
Class D Switching Amplifiers
I purchased a 4X100W switching amplifier board made by Sure and sold by Parts Express. It has decent specs and has fairly quiet audio. Amazing deal at $48. Quiescent current is lower than my LM3886 amps, and 2 of these boards will stack in about the same space as my 2 x 4channel amps. It hasn't gotten past the lab bench though. I suspect that crosstalk will be a problem, just as it is on any multi-channel amp unless it uses the input ground fix I describe above. On a switching amplifier this problem is harder to solve, I suspect.
Sure 4x100W
Photo from Sure

2/1/12 Yet Another Audio Amp
SparkFun has a stereo amp kit based on the ST540 Stereo Amplifier chip. This chip is a good quality 35W per channel amp. It has 2 channels per IC, uses a single 20V power supply (bridged outputs) and has a nice shut-down feature, unlike the LM3886. It also has good stability when driving long (capacitive) speaker cables, requiring no Zobel (RC) networks. Driving long speaker wires is key for a whole-house stereo. SparkFun gets $4.95 per chip, qty 1, $3.96 for qty 100, much less than two LM3886s. I think I can easily build 3 x 2 channels on the same ExpressPCB Mini board size that I used for four the LM3886s. The design will require a similar grounding trick to the one used on the LM3886 board to prevent channel-to-channel crosstalk.

Sparkfun STM540 Amp
Photo from SparkFun

2/16/2013
Web page??

I want to add a simple web page to control the system. Since the serial port has commands to replicate all the buttons, that is a good starting point. I'm no networking guru, so I'm looking for a simple solution to adding Ethernet. There are lots of ways to get networking on an STM32 including these that I have considered:
  1. Use a STM32 processor with Ethernet (STM32F107 processor plus an external PHY and mag-jack)
  2. Use a Wiznet like W5100 (SPI to 100BaseT) chip or module
  3. Use a Serial to Ethernet device. XPort looks good
  4. Use an external Linux CPU with a proper web server and serial.
  5. Use Arduino and an Ethernet Shield
1. would require a new CPU board. I would need to buy a $300 eval board to get started, then either use the eval board in the product (expensive) or build a CPU with Ethernet PHY. I still haven't seen a cheap STM32 board with Ethernet. Nor have I  found a simple HTTP example for the STM32.

2. could use a Wiznet SPI to Ethernet module such as WIZ810 ($20)on my current board. I haven't seen any port of the Wiznet code to STM32, but it probably isn't too big a job. Where the STM family has a version with built-in Ethernet, this is not the most elegant solution. The problem with Wiznet is its HTML code is pretty rudimentary and after that I'm on my own. Like I say, I'm no networking guru.

3. Lantronix XPort is $45 and is packaged in a very small, extended-length RJ45 connector. It comes with HTML, FTP, Telnet, TCP, Email, decent tools, and is pretty easy to use. It has a Java run-time engine (JRE) so you can install Java code to build GUIs and applications. Java is probably a decent skill to have The Lantronix tool for downloading code is a bit clunky, requiring you to break your filed into chunks that fit in 64K. Clearly a proprietary, closed solution.  

4. is promising. Embedded Linux is good where you need access to networking, server, USB, video, audio, big GUI, file system... I'm not seeing any ARM Cortex M3 Linux ports since Linux needs too much memory for M3, requiring large chunks of SDRAM and flash. Until recently an embedded Linux board was about $100 but there are sub $50 boards available. Raspberry Pi claims impressive specs and capability at a $40 price tag, but it is a bit of a toy. It is not open hardware and not available for commercial design, since the processor is available for high volume (consumer) use only.  Olimex iMX233-OLinuXino-MAXI is an open hardware solution and looks decent for $60. If I were to want to add audio server (audio file player, ether network or local thumb drive, Pandora or on-line music, etc) capability to this project, Linux would be the way to go.

5. the Arduino approach is appealing since there are both  hard-wired Ethernet and WIFI shields available along with decent web server demos. Two possibilities:
    1) add a low cost Arduino (Leonardo + Ethernet Shield) to the STM32 as a simple stand-alone web server, or
    2) use a higher-end Arduino Due (Arm based) and eliminate the STM32.
Option 1) is probably quicker and easier and would be a good starting point if I ever decide to move to 2).  It has the disadvantage of adding 2 boards at a cost of about $70.

Tools update
In case you wondered how ST can sell the Discovery series of development boards for $10 and up, the answer has arrived. You can only purchase them in small quantities, but only for development. The Discovery modules are not to be used for a product. I'd like to see someone build a low cost STM32 module.

On the compiler front, Atollic further restricted their free compiler version. While I was developing code, they dropped the number of code breakpoints from 3 to 1. That wasn't terrible, just an inconvenience. But in 2012, they dropped the maximum code size from 128K to 32K. The full version of Atollic is quite expensive.  Since my code size is 48K and growing, that is a deal breaker. I installed CooCox ARM compiler suite and with minor code cleanup, got it to compile nicely. CooCox is a very well done Chinese development tool based on gcc and Eclipse IDE. It supports 9 manufacturer's ARM chips, hundreds of chip versions, and many debugger choices. Truly great!

12/16/2013 
Circuit Cellar Magazine Article

In the spring of 2013, I contacted Circuit Cellar about publishing an article for this project. I wrote it, they accepted it and it appears as a two part article in the Jan and Feb 2014 issues. Here are .PDFs of the Jan and Feb articles. All of the design files are available on the Circuit Cellar web site as well as here. Let me know if you are interested in building one. I am considering offering bare or assembled PC boards. Email me at dave@djerickson.com

7/2022  Update
I have been using the Whole-House audio system daily, and pretty much unchanged for the past 9 years. One of the input sources is connected to a Bluetooth adapter. I also built a music server using a Raspberry Pi running Volumio software. This is a great way to access my music library, on-line music, and web radio stations with a nice web control, and without requiring a PC.

After considering the above ways to implement a remote web page control for the Whole-House system, several newer technologies have become available in the IOT market. I am looking at an ESP32 and MIT's  App Inventor. This looks like a minimum learning curve solution. The ESP32 will convert commands from web based messages to the serial commands that the system uses. I haven't done it yet, so wish me luck. 

I no longer bother with the old 128 x 64 LCDs and their parallel bus. The modern 128 x 64 OLEDs and QGVA TFTs are nicer, use simple control over SPI or I2C, and the libraries such as U8G2 are very nice. Speaking of which, I am considering redesigning the Whole House system to use an Arduino processor instead of the STM32 Discovery.

4/2023 Update

Finally built a web page control for this system, using an esp32.The esp32 serves up a simple web page of 20 buttons. It requires a bit of HTML, CSS and some Javascript. Fortunately I found a nice example to use. Check it out!


webpage

STM32 Page
128 x 64 LCD Page
esp32 Controls for Multizone Page
   Dave's Home Page

Last Updated: 4/23/2023