Duration: April 2002 – December 2003 (Part Time)
Position: Design Engineer
Developed 120W, 250W and 150W linear audio amplifiers.
Worked on 3kW Class D amplifier and DSP for active sub-woofer application.
Developed short range RF remote control for audio visual equipment applications.
Digital signal processing algorithms for speaker frequency response correction and room response correction.
Active cross-over network for Loud-speakers.
3kW prototype Switch-mode power supply.
The partners of GGAcoustics wanted to develop their own brand of high-end loudspeakers and audio equipment. They were very happy with the performance of the 120W amplifier I had developed in my own time with their funding (see personal projects – 120W Amplifier), but they wanted something a bit more powerful to use as a reference amplifier for testing loudspeakers. So my first task was to rework the 120W design into a 250W design. This was fairly straight forward as it was a repeat of the design process I used for the 120W design. I made use of the same transformer simulator application as mentioned in my personal projects section to arrive at a specification for the transformers. I modified the existing PC board design for the 120W amplifier to accommodate TO-3P power devices instead of TO-3 devices and there were some other minor modifications and improvements to the previous layout as always happens when one reworks a PC board. The only major difference in the design was the power supply voltages and voltage head room. I constructed the amplifier in a similar fashion to the 120W design mounting each board on a single large heat sink and then mounting the amplifier modules, transformers, capacitors and soft start on a large piece of wood.
Sonically the design was slightly better than the 120W amplifier, possibly due to the use of different output devices and some other minor changes. The partners were happy with its performance. Although they both owned Mark Levinson No 336 amplifiers, the No 336's were starting to fall out of favour with them in preference to my 250W design and eventually they both sold their No 336's.
My next task was to design a high power class-D amplifier with DSP filtering for an active sub-woofer prototype. Although the ultimate intention was that we would design all the circuitry ourselves for this sub-woofer, initially we made use of some OEM 1kW modules to drive the sub-woofer for testing purposes whilst I got busy with designing a PC board for the DSP and analog circuitry. I made use of the Freescale DSP56F800 device – not a particularly high performance DSP, but for use in a sub-woofer it was adequate. Having laid out the PC board, getting it manufactured and then stuffing it myself I got busy with the DSP software making use of Freescale's CodeWarrior IDE and compiler for the DSP56F800 devices.
The software requirement was to be able to provide an inverse filter response to correct the sub-woofer drivers frequency response to near flat and then below 20Hz provide a sharp cut-off to limit driver excursions. In addition it was desired to be able to load any custom frequency response into the DSP over and above the driver correction response to perform room correction. The filter requirement called for the use of a minimal phase FIR filter kernel and I spent a fair amount of time researching how to convert a symmetrical linear phase filter kernel into a minimal phase kernel.
After getting the DSP software working to a point where it would at least correct the sub-woofer drivers to a flat response and could be used for further testing I concentrated on working on the Class-D amplifier. We had purchased a number of monolithic Class-D modules from a company called Tri-path which we hoped to use for the Class-D amplifier. These Tri-path modules provided the PWM modulator and driver circuit and required external MOSFETS for the output. I designed a custom PC board to mount the Tri-path module, MOSFETs, output filter and associated circuitry. I used some large toroidal transformers for the power supply, making use of my transformer simulation utility to select them. Although the board functioned reliably at lower output power levels (less than 500 Watt), we experienced numerous MOSFET failures trying to achieve higher output powers. Even after a second iteration of the board design, the failures continued. This was possibly due to my then lack of experience in good layout techniques for Class-D and switching MOSFET designs (hey, one has to learn somewhere!), but also due to limitations in drive capability of the Tri-path modules themselves.
Alongside the Class-D development I also worked on an RF remote control and display card combination. The Display card was intended to be used with the active sub-woofer for the user interface and it would be controlled by the RF remote control. The GGAcoustics partners preferred the use of an RF remote instead of Infra-Red remote since it would have better range, be more reliable and not require pointing the transmitter directly at the receiver window. Thus the receiving equipment could be hidden from view for aesthetic purposes. Performing remote control using RF does have certain complications which do not apply IR remotes, for example the transmitter and receivers need some sort of additional coding to pair them such that one transmitter will not affect all other receivers within range.
I made use of the Chipcon (now Texas Instruments) CC1000 short range ISM band RF transceivers for the remote and display card. A very nice feature of these devices is that they provide the data slicing and clock recovery internally, thus simplifying the software algorithms used to control them. When placed in receiving mode they output a clock and data signal at logic levels and the user must write software to detect transmissions and lock the internal data decision filter. In transmitting mode the CC1000 provides the clock signal and the user must supply the data stream synchronized to the CC1000 clock. So when writing a software driver for this device one can use processor interrupts initiated by the CC1000 clock signal in both transmit and receive modes.
I made use of a Freescale HC08 microprocessors in both the remote control and display card. A remote control device implies battery operation and long battery life, so I designed the remote control to work off 2 AA Alkaline cells - hence 3V supply and made use of the low power standby modes of the microprocessor and CC1000. There was also a small amount of other logic circuitry required to provide the multiplexing for the remote buttons. When all devices on the remote control were put into standby mode the board drew less than 1uA. The display card consisted of the CC1000, microprocessor and an Osram dot matrix LED display. These displays are fairly typical in consumer audio visual equipment. I used PCB loop antennas on both the remote control PC board and display card board for transmission and reception of the RF signal.
From a software point of view, using two processors from the same family meant I could write one driver for the CC1000 that would work on both the remote control and display card and then simply set up some port pin defines in a header file to keep the driver module itself platform independent. I also added a software serial driver which likewise could be used on both platforms. To implement the pairing requirement I included a unique serial number in the data protocol. The transmitted packet consisted of a length of preamble to train the bit decision filter on the receiver, a frame sync byte, the unique serial number, a size field and data field followed by a CRC field. The receiver algorithm not only had to parse this data and perform the CRC check, but it also had to drop out the algorithm and be ready to start again if an error was detected in the data. The reason for this being that when the CC1000 is in receive mode it is constantly streaming a clock and data signal even in the presence of no transmission so there is a small but real probability that the receiving algorithms can incorrectly detect data transmissions from the noise. Getting the algorithms to be really robust in these circumstances took a lot of effort and experimentation with different length preambles and data packets. However I was ultimately able to achieve similar performance to IR remotes with no random false detections from the noise. Finally on top of the lower level data transmission and reception modules I wrote a basic menu system for the display card which responded to key commands from the remote.
Other tasks I performed for GGAcoustics involved constructing various active filter circuits for cross-over networks. I also began a new linear amplifier design in the form of a 150W amplifier. The partners wanted a high audio quality compact design that could be used in amplified speakers. So for this design I migrated all components of my existing amplifier designs to SMD where possible. I followed the same design process as with my previous amplifier designs. I had some boards manufactured for this amplifier and assembled and tested them, but did not complete the system whilst I was still working for GGAcoustics.
The final task I undertook for GGAcoustics was to develop a 3kW prototype off-line switch mode power supply. This supply was intended to be used for the active sub-woofer. I used an H-bridge configuration for the MOSFETs and the MOSFETs were driven by an SG3524 PWM controller using two International Rectifier IR2113 high and low side drivers. Getting a such a high power switch mode to work is no easy task. There are large capacitances around the power devices which require strong drive capability from the H-bridge driver. In South Africa getting hold of specialist devices in small quantities for prototyping can be quite difficult and as such the only MOSFETs I had available had very large gate capacitance. They were a bit of an overkill for this application. Thus I had problems with the IR2113 drivers overheating. In addition in an H-bridge configuration when one pair of MOSFETs switches off and the other pair turns on a large charge needs to be shifted to get the voltage polarity across the transformer to switch. By careful design of the transformer this effort can be removed from the H-bridge drivers and placed on the transformer. I was still working on these issues towards the end of 2003 when the partners of GGAcoustics decided to pull the plug on the development of electronic audio equipment and focus purely on loudspeaker development. Getting an engineering development and manufacturing business off the ground is a very costly venture and they had decided to curtail some of the development expenses. Thus I ceased working part time for GGAcoustics and became full time employed by Natcom Electronics. The partners however allowed me to keep some of the electronic items I had developed for my own use and I have since remained in contact with them. They subsequently launched the company name Vivid Audio.