8. Case study #2: a 3-way design


Case study #2 is a 3-way plate amp designed for stereo satellite speakers that includes a WiFi adapter, programmable DSP, and 6 amplifiers for plenty of power.  It is intended to be a DIY version of the new breed of wireless speakers such as the Sono, Heos or Bose systems, but with higher quality drivers and more opportunities for the DIY’er to customize the system.

This approach puts all of the electronics for a stereo 3-way system onto a plate amp that can be mounted in the subwoofer.  The subwoofer provides the low frequencies for the 2-way active speakers.  The mid-woofers and tweeters are in separate boxes that can be placed on the desktop or on either side of a TV.  But this is just one possible implementation:  this WiFi/DSP/amplifier could also be used for a “true” active 3-way system, where the electronics are in a conventional chassis instead of on a plate in the subwoofer.

The electronics includes the 3-way DSP board described in article #6, along with a commercial 6-channel power amp, the power supply, and a WiFi module for the audio source.  There is also a lot of software that goes along with this design, and that is addressed in article #12.  This article focuses on the design and construction of the plate amp and how the various components interconnect.  Article #5 describes the subwoofer box that the plate fits into, as well as the software for the boost and bass enhancement.

Much of this design was originally done as a Parts Express (PE) Speaker Building Design Team (SBDT) project, and pieces of this project are described in write-ups on the PE Project Gallery.  The plate amp project is titled “Act 2” in the PE Project Gallery, and the surround speakers are described in the project titled “Act 3”.  The use of the term “Act” was an inside joke, about this design being a DIY version of the Sonos Play speakers, and it took 4 acts to complete the play.

Part 1:  The Plate Amp

The block diagram for the plate amp is shown below, and the next sections describe the various components and why they were chosen.


As you can see from the block diagram, there is a lot happening on the DSP/CPU board. I used a custom design for this board to keep the circuitry compact and avoid wiring between submodules. The CPU, DSP, DAC, local power supplies, and Bluetooth/USB are connected by traces on the board, and the only external connections are the audio lines (which use RCA phone jacks), power and some serial connections.  This board is described in detail in article #6, and a newer version of the board that will be easier to build will be released soon.

It is possible to implement the DSP/CPU with separate components, such as a miniDSP or FreeDSP board, an Arduino CPU board, a DAC board, and a Bluetooth module.  But wiring can be difficult, and at some point it makes more sense to assemble a custom DSP board than to spend even more hours trying to wire up all those modules and find ways to secure them.  But it’s certainly not impossible to use separate pre-built modules for the DSP/CPU functionality:  that’s just not the approach described in this article.

Amplifier and Power Supply Selection

The design goals include a subwoofer plus left and right active 2-way speakers, so I needed an amplifier with a minimum of 5 channels (subwoofer, left woofer, left tweeter, right woofer and right tweeter). Since the Act 1 subwoofer described in article #7 uses 2 drivers, a 6-channel amp was ideal. I didn’t want to deal with heat build-up so I only looked at class D amplifiers. Class D amplifiers switch at high rates and the switching currents and magnetic fields in the output inductors can cause problems when trying to use multiple two-amplifiers near each other. So, the logical selection was the 6-channel Sure amplifier, PE part number 320-307. This amp uses three of the TDA7498 chips synchronized to the same clock, so the amps will not interfere with each other.

The TDA7498 is a well-designed class-D amplifier IC, and we can count on reasonably good audio quality from this amp. The amplifier comes with a small fan to blow air across the heat sink. You will need to remove the fan, as we are going to mount the amplifier “upside-down” and thermally connect the heat sink to the mounting plate. To ensure a good thermal connection, use a .5mm thermal pad between the mounting plate and the heat sink. When you use the 20mm standoffs in the kit mentioned below, there will be enough pressure from the mounting screws to provide a good thermal connection. There is a picture that shows the blue-green thermal pad between the mounting plate and the heat sink.
The power supply is the Meanwell 150W 24V switcher for the larger plate amp, and for the smaller plate you will need just about any competent 100W-150W external “brick” type power supply. A larger power supply could be used, but a nice feature of active 3-way systems is that you can typically get by with smaller amplifiers and power supplies than for similar passive systems. The 100W to 160W power supplies work very well for this application.

Unfortunately, like many of these small amplifier modules, the output filters use ceramic capacitors that introduce some distortion.  If you are ambitious and want to improve the sound quality a bit, you can change the output capacitors to film types.  It’s not hard to do, but it takes a bit of time and it helps to have an SMD rework station.  It will also void your warranty, so make sure you put enough thought into this decision.  The picture shows the yellow film capacitors that I used to replace the output capacitors.  Professor Trevor Marshall provides some measurements of the Sure TDA7498 amplifier with both the supplied ceramic capacitors and the replacement film capacitors, and the result is a 10-fold reduction in distortion in the range above 1KHz.  If you are going to be using very high-resolution drivers for this WiFi system, I recommend changing the capacitors in the output filter to film types.  You can see the film capacitors that I used in the picture above, on the left side of the picture.  However, for many applications this extra step is probably overkill.  The picture also shows that I moved the speaker connectors to the bottom of the board to make them easier to get to.

Mounting Plate

The subwoofer cabinet used in article #7 is the B-REX subwoofer cabinet, which has a removable back panel.  The plate amplifier replaces that back panel.  Because that panel is relatively large, I used a ¼” aluminum plate to ensure that it would not vibrate. Just about any grade of aluminum is acceptable, but I used a sheet of 5052-H32 cut to 12-1/2 by 12-1/2 inches from onlinemetals.com. You can cut aluminum plate on a bandsaw, but it’s so much more convenient to purchase the material already cut to the proper size. Whatever you do, don’t try cutting the plate on a table saw—it is dangerous. Aluminum tends to kick back very easily on a table saw, and flying aluminum is dangerous and potentially lethal. Onlinemetals often has promotions and special deals on shipping so you might want to combine your order with a friend to qualify for one of their special deals. One of the ¼” plates cut to size should cost somewhere between $20 and $35. For smaller mounting plates, the plate can be thinner.  There is an alternate version of the plate amp designed to replace an SD100 amp, where the longest dimension is only 8”. For that I purchased a 3/16” thick panel, and the price was around $10 per plate.

The specified dimensions for the back panel of the B-REX enclosure aren’t accurate, at least for the box I received.  The specification sheet lists it as 319.6mm, which is 12.58 inches. However, the opening measured a slight bit under 12.5” on my cabinet. I had to trim down the 12.5 by 12.5 plate using a 12” disc sander, although a belt sander would have worked just as well. Fortunately, aluminum sands very easily and it doesn’t take long to shave off the small amount needed to make it fit. I used the old back panel as a template for the screw holes, and even though the panel is ½” thinner, the old screws worked OK to secure the aluminum plate to the cabinet.

The ¼” thickness of the plate means that you will need lots of long screws with standoffs to attach the various boards, or else you will need to countersink some shorter ones. I went with countersinking. PE offers a nice kit of screws and standoffs that make mounting hardware to the plate a lot easier. The kit is part number 320-3287, which has an assortment of hex head screws to go with the standoffs. Standoffs are somewhat expensive from places like McMaster-Carr, but the Sure/PE kit is reasonably priced. Hex head screws are easy to countersink and look nice when you countersink to a depth that makes them flush with the mounting plate.

WiFi Module

Parts Express carries two different WiFi modules—the WFA02 (#300-576) and the WFA28 (#300-577).  I used the smaller (and cheaper) WFA02 module.  Both are controllable with the same smart phone app, and both offer the features we need for this project.  The functionality of this module is described in the article on WiFi audio.

Plate Amp Layout

The first step in building any plate amp is to lay out the parts so they fit on the plate. This can be done by using “paper dolls” of each component or using a computer drawing program. I used Powerpoint to create the components using a 2:1 scale, and moved them around on the plate to make sure everything would fit. The Act 2 plate layout graphic shows that the parts fit with little room to spare. There is also a layout for a comparable plate amp using the Sure DSP boards. Even with the extra modules, there is enough room on the plate for everything. But with the Sure modules, there isn’t enough room to use the bulky RCA connectors as interconnects, so the cabling will be more involved.

The SD100 replacement amp is much smaller than the 12.5” square plate amp, and it is clear from the layout that the power supply will need to be external. And even without the power supply, the 7” by 8” plate needs a second layer to make sure the components fit. I used a 1/8” thick plate of aluminum that I had left over from my wife’s business. The guys in the machine shop where I worked had made her some nice aluminum cookie sheets in exchange for some cookies, and I was able to cut the cookie sheets down to the right size on a bandsaw. So, if you don’t have extra aluminum plates laying around, you will need 2 pieces of 6” by 6” 1/8” material. The picture below shows the double-decker construction for the SD100-sized amp.


It’s important to drill all the holes you need to mount components before painting and assembling the parts. Drilling is messy, and you don’t want metal flakes shorting out circuitry that is already mounted. The hole locations and drill bit sizes and countersinking will vary depending on the selected connector hardware and the modules you use. Take your time on this important step, and keep double-checking for the right fit for each item. Once the holes are drilled, you can sand the plate with increasing fine paper until you get the “texture” you desire. You don’t need to paint the aluminum plate, but a coat or two of flat black will be less conspicuous and easier to get past your wife.

The power supply has pressed-in threaded inserts for mounting that are difficult to accurately locate. Fortunately, the power supply has a plastic insulator that easily slides out after removing one screw located near the power input connector. Just mark the insulator with a pencil through the mounting holes and then slide out the insulator sheet to use as a template. For the DSP board, just make a printout of the PCB or one of the Gerber files to locate the mounting holes. Once you drill these holes for the M3 screws (1/8” will give a tight fit), countersink the cap screws with a 5/32” drill to a depth of 1/8” from the front side of the plate.

The Speakon and Powercon connectors require an “oddball” 24mm drill size. Since this is not a common size for most drill bit kits, you may have to order one online. I bought a 24mm “Roman Carbide” Forstner bit from Amazon for $13. However, after drilling just 2 holes in the aluminum plate, one of the carbide blades broke off. By the time I was done with all the holes, both carbide blades had separated from the body. You might want to either buy a better drill bit or else be very careful to not press too hard with the budget drill bit.

Notice that many of the components on the plate require countersinking from the back due to the thickness of the plate. You might be able to find some panel-mounted switches and connectors suitable for ¼” panels, but you will probably have a better selection of components if you just countersink everything with a large bit.  And if you are building the SD100 version, you might need to shop around for a heavy-duty power connector. A lot of the 5.5mm/2.5mm barrel connectors are only rated at 3A at 12V, but Digikey has one that is rated for 5A at 24V.

You don’t need mounting holes for the WiFi module, as that plastic box can be affixed with Velcro or the 3M mounting strips used for toll road transponders.

Assembly and Wiring

Assembling the components to the plate is fun, as you get to see the amp take shape. But wiring seemingly takes forever, and there is little joy in crimping pins on small ribbon cables or soldering RCA connectors to shielded cables. Plan to fight the tedium by having good beer or stronger spirits to keep you amused, and don’t get frustrated at the amount of time it takes—that’s just the nature of this task.

A step that requires some extra care is wiring up the external antenna to the WiFi module. You only need to do this if you can’t get a good WiFi signal with the module hidden behind that mounting plate. If you are close to your WiFi router you will probably get an adequate signal. But our router is in the basement, so I wanted an external antenna, even though I knew this would void the warranty. Carefully open the WiFi module using a knife blade to pry off the cover. Then remove the 4 mounting screws holding the board to the plastic housing. The button for the WPS switch is probably going to fall out, so make sure you don’t lose it on the floor. Then remove the antenna connector—it is one of those tiny U.FL connectors that you see in laptops, and it’s got hot melt glue that you need to pull off. I used a U.FL to RP-SMA cable assembly (Mouser part #741-080-0001) to allow using a nice external antenna. With the external antenna, the signal strength was about 98%, even though I was two floors up from the router. If you have a hot melt glue gun available, put a small blob of glue on that UFL connector to make sure it doesn’t fall off the board (that happened to me!). When you put the board back in the case, make sure you remember the WPS button, because you will need it to get the WiFi module to join your network.

Unfortunately, the original wiring approach had issues with grounding. The WiFi module requires +5V at around 300ma, and most of that current is for the WiFi signal, which in effect is a small radio station. When you try to connect the audio to the amplifier or DSP, all that return current is flowing through the ground, and it creates an audible noise like digital chatter that is annoying. It would be nice if the WiFi module had separate analog and digital grounds that could be connected independently, but that’s not the case. There may be other ways to fix this grounding issue, but the easiest approach I found was to either use a DC to DC isolating convertor for the WiFi 5V, or else run the WiFi from a USB charging module. The noise goes away once the power is isolated. The new version of the DSP board will include the DC-DC isolation module, but for this prototype I used the USB charger.

Loose Ends

This project requires loading software into the CPU. Unfortunately, software is never done, so it is helpful to extend the USB connection to the CPU by adding a connector to the outside of the box. The Neutrik NAUSB-W-B (PE part #092-279) and a short micro USB to USB cable (PE #130-572) let you easily provide an airtight USB connector on the subwoofer front baffle. There’s not enough space on the back panel using the layout shown here, but having the connector on the front is a convenience.

Another important tip is to make sure you set the switches on amplifier that adjust the gain before you mount it to the plate. Since the amplifier board gets mounted “up-side-down”, these switches will not be accessible once the amp is installed. There are some calculations that you can do to determine the best gain value for each channel, based on the power supply voltage and the sensitivity of each driver (see table below). For many designs, the best switch settings will be “weak” (25.6dB) for the tweeter and subwoofer and “low” (31.6dB) for the woofer or midrange.

Part 2:  The Surround speakers

The plate amp gives us a powerful DSP processor, 6 channels of amplification, and 24-bit lossless audio via WiFi.  We still need some speakers to connect to this plate amp and in this section we will look at the requirements for the surround speakers to understand our options.


The DSP on the plate amp greatly simplifies the surround speaker design. With the DSP and separate amps for each driver, we don’t need to design a passive crossover, and we have more options for sizing the enclosure.  The DSP uses the popular ADAU1701 chip, but with a local CPU host that can calculate the filter coefficients from simple commands sent from a cell phone. So, if we change drivers or crossover points, our generic design can be tweaked and optimized by simply sending some commands from our cell phone.

There are many benefits from using DSP and dedicated power amps for each driver, but the main benefit is that it simplifies the crossover design. Specifically, the designer doesn’t need to deal with the complexity of passive crossover design. Passive crossovers interact with the driver impedance, so the designer needs to have reliable impedance measurements. Since the impedance of a typical driver changes with frequency, and because many drivers have different impedance curves, the passive crossover network needs to be designed for a specific set of drivers, and will only work well with those drivers. For an active system, the driver impedance curve is irrelevant, as the low impedance output of the amplifier completely “swamps out” the driver impedance. We still should measure the amplitude response of the driver, but for many drivers that are “well behaved” in the intended frequency range, even the amplitude response of the individual drivers can often be ignored.

The DSP crossover for the woofer and tweeter can be selected from a simple menu that currently includes 5 different crossover types (one-pole Butterworth; 3-pole Butterworth; 2-pole Linkwitz-Riley, 4-pole Linkwitz-Riley, and 8-pole Linkwitz-Riley). The crossover frequencies can be selected from 7 different frequencies (1200Hz, 1600Hz, 2000Hz, 2400Hz, 2800Hz, 3200Hz, or 3600Hz). Similarly, the subwoofer to woofer crossover can be selected from 4 different types and 5 different frequencies. These menu options are in a look-up table, and could easily be changed by either modifying the code or downloading a custom table.

The DSP also provides baffle step compensation, selectable from 8 different frequencies ranging from 300Hz to 1000Hz, and 7 different levels from 0 to 6dB. Additionally, there is a 9-band equalizer that can be used to tailor the frequency response. And the volume of each driver can be adjusted over a 10dB range.

These DSP functions give us a lot of control over the drivers, and the settings are saved in non-volatile memory (EEPROM). So, we can optimize the settings for a given set of drivers and save those settings, and recall them later. By using a standard connector for the satellite speakers, we can quickly connect different cabinets that have different drivers, and simply recall the optimize settings for each speaker. There is a lot more discussion about the software and DSP control in article #12 of this series.

Drivers and Cabinets

One of the nice benefits of using separate amps and DSP in active speakers is that driver selection tends to be a lot less critical than with passive designs. If you use drivers that behave reasonably well in their assigned frequency band, you can usually use them successfully in an active speaker, even though they may be difficult to work with in a passive design. So, this was an ideal opportunity to draw on my stash of drivers in the attic and put some of those unused Christmas presents to good use.  Also, because the volume of the cabinets were in the 4 to 6L range, I was able to build them using solid hardwoods rather than MDF.  Hardwoods can have issues with cracks and splitting for cabinets that have large surface areas, but for this volume, hardwood cabinets are usually successful.  And the reason for wanting to use hardwood cabinets should be obvious:  they look nice.

I built two sets of cabinets that use the same type of connector, to allow comparing the surround speaker designs.  One of the cabinets used locally-grown cherry from our cabin in Western Maryland. This wood has a large amount of figure, so it would be classified as curly cherry, yet it is very stable and easy to machine. When used with a penetrating finish such as tung oil or boiled linseed oil, the wood takes on a rich depth that creates refraction patterns of dark and light that move as you change viewing position. This interesting and visually appealing scattering of light is called chatoyance.

These cherry cabinets used some SEAS Excel W14CY-001 woofers purchased many years ago, along with the Tymphany (Vifa) NE25VTA-04 aluminum tweeters. The Excel woofer uses a magnesium cone with prominent breakup modes in the 10KHz area, and it has a well-deserved reputation for being a “difficult” driver to work with. However, a steep 8-pole crossover at 2KHz or so will completely suppress those breakup artifacts. And this relatively low crossover point is not a problem for the ¾” Vifa tweeter, especially with that steep filter.

The second cabinet used a plank of zebrawood purchased for a bass guitar body project that never happened.  The  tweeter selection for the zebrawood cabinet was one of those ideas that gets stuck in your head and won’t leave you alone. Every time I looked at those zebrawood stripes, I was reminded of the AMT drivers that PE is carrying—the AMT mini-8 and the AMT-2. However, neither of those tweeters was “ideal”, as the mini-8 didn’t extend low enough to integrate well with a 4” or 5” woofer, and the AMT-2 suffers from some “beaming” for frequencies over 5KHz due to the relatively large size of the diaphragm. So, I decide to use both tweeters, with a passive crossover at 5K between the two. This passive crossover was a simple 2-pole Butterworth filter. Because the AMT impedance is almost entirely resistive, I could use textbook values for this crossover.  The woofers for the zebrawood cabinets use the MCM 55-5655 driver.  This driver was the right size for the cabinet and it looked like it would be a good performer, although it now appears to be unavailable.

Cabinet Construction

Both cabinets were sealed, in the 4L to 5L range. This internal volume results in an F3 of 90Hz to 120Hz for the drivers that were considered for these cabinets. Since these will be used with a subwoofer and an active crossover for the subwoofer to woofer transition, the box dimensions are not critical. In fact, I simply used the amount of wood I had to determine the box final dimensions, and just scaled everything to achieve a “golden ratio” (1.6) of height to width.

As noted previously, one of the motivations for using hardwoods is that they look nice.  And one of the best ways to make it obvious that they are solid hardwood is to use interlocking joints for the corners, to show off the end grain.  I used dovetail joints for the cherry and box joints for the zebrawood.  These interlocking joints are very easy to do with a dovetail jig–I used the Porter Cable tool that runs for about $100 at Home Depot.

Dovetail joints are amazingly strong, and are overkill for ¾” wood.  But, as the pictures show, they look nice.


Voicing and Testing

You can determine the optimal crossover slope and frequency using the modeling and measurement techniques that work in a similar way for passive crossover design. It’s still important to make sure cone resonances are not excited and that areas of high distortion are not allowed in the crossover passband. The only effective way of meeting these goals is to look at the individual driver response and the crossover response and summing their transfer functions. This can be done by modeling programs such as Active Speaker Designer (Audiodevelopers), Active Crossover Design (Charlie Laub’s spreadsheet), or Soundeasy (Bodzio Software).

But one of the nice benefits of using active crossovers is that with “well behaved” drivers there is a high likelihood of having a good sounding system with some simple tweaking of volume levels, BSC and delay, and intelligent selection of the crossover parameters. You should still measure the speakers to make sure the volume levels are properly set and that the phase is correct for the selected crossover (there is a polarity button on the cell phone app). But oftentimes those complex modeling tools are not needed for good results.

The cellphone app that controls the DSP parameters is a convenience, although once the speakers are properly voiced, there shouldn’t be a need to ever change the settings.

Final Notes

This a big project that brings together many of the pieces that have described in previous articles.  It’s still a work in progress, as we will likely make some additional changes to the DSP that allow more control over the final response and add bass enhancement algorithms that might allow reducing the overall size.  We’ve brought together the technologies to make very high quality speakers that sound good and look good, and we have a nice test bed to continue our experiments.