The Teensy 4.0 and 4.1 processors are widely used for audio projects. To this end, PJRC made an Audio Adaptor available that uses the SGTL5000 stereo (2-channel)) CODEC IC (ADC and DAC combo). A study of this device and a comparison with the TI (Burr-Brown) PCM1808 ADC showed some significant advantages for the 1808 (more below). This PC Board allows easy useage of the TI parts and also provides four channels on the same physical format as contained the two-channel SGTL5000. This page is about the construction of the board. A separate page compares the two CODEC options with measured data.
This 4-layer PCB is available from OSHPark in multiples of three. There are no part kits, but all parts are available from Mouser and others.
This board was constrained by having the same size and connections as the original SGTL5000 board. That dictated that a four-layer PCB be used and even at that, two mounting holes were removed to gain space. The four IC's use the older TSSOP packages that require more space than some of the latest designs. That asks for more space on the PCB, but with a huge advantage of being hand solderable. Most of the other components are 0805 with a few 1206 size inductors and filter capacitors. All of those have hand-solderable pad laypouts. There are 28 pins with 0.1 inch (2.54 mm) spacing that line up with either a Teensy 4.0 or 4.1 processor. Note that the pin layout is different for the earlier Teensy 3.x processors, but the board is electrically compatible.
The circuit - Both of the IC types are programmed entirely by hardware pin connections. This means that there is no serial programming connection, like we are used to. Bot devices have some special needs for power filtering. The analog supplies need to be reasonably clean of digital noise and here we include 1 mH chip inductors along with 100 uF chaip capacitors. The 5102 DAC was designed to have over 5-Volts p-p output swing that is centered about zero Volts. This requires a couple more large value chip capacitors. There is some small "alias" noise on the DAC output that is in the MHz frequency range. That is filtered with an RC network. The schematic below tells more details.
This board was a continuation of experiments in power filtering. It uses more series inductors on power line than are usually seen. There are no paralleled unequal vales of capacitance used and the high capacity MLCC have been used throughout. See this note for more information on these ideas. Initial measurements on this board have been favorable.
The board - All parts are mounted on the top side. The silk screen markings were cut in some places. The four layer board uses layers 1, 2 and 4 for circuit traces and layer 3 is part analog ground and part digital ground. The two are joined at a single point by a zero-Ohm resistor, R17. If two DAC ICs are used, it is necessary to run a wire from J8 on the CODEC board to pin 32 on the Teensy 4.x board (that pin is not available in the 28 pin connections). In the case of the Teensy 4.0, pin 32 is accessed on the back side.
Information about the PCB is in a single ZIP file, with all Kicad files, including the Gerber zip file.
Here are a pair of BOMs for Stereo and 4-channel. Sorry it is not text, but this is what was easy as well as being easily read.
Using the board - This is not a drop-in substitute for the SGTL5000. It does not have a number of the SGTL5000 features. For instance, there is no microphone hardware nor separate audio path, there is no headphone amplifier nor jack, there is no SD card nor Flash/RAM memory pads. But, there are four independent audio inputs and outputs and the dynamic range of the ADC is considerably better. See this comparison for more information. The PCM1808 also has no (or very low level) idle tones.
The example INO files for the Teensy Quad I2S can be used as is. Or all reference to SGTL5000 can be removed, as there is no I2C hardware to be controlled. No Codec needs to be referenced in the class instantations. Additionally, here is an .INO file to test this board. It is the minimum that tests all four in and out channels.
When used with the Teensy I16 library, the PCM1808 ADC will fill the data word with significant data and the noise level will bw partly suppressed, altering the answers. If used with the F32 library, the 24-bit mantissa will completely represent the data and noise and is measurably better for extracting weak coherent signals. Depending on the application, this may or may not be important. As of this July 2025 writing, there is only 2-channel I2S-input for the F32 library. More later as this is improved on.
The spec sheet for the PCM1808 ADC says, "Sampling Rate: 8 kHz–96 kHz" and the Clock Frequency chart lists various sample rates from 8 to 96 kHz. Non-audio applications, such as radios and instrumentation, can benefit from using higher sampling rates. So, I explored the idea of higher rates by using the T4.x rate changer routine (see https://forum.pjrc.com/index.php?threads/change-sample-rate-for-teensy-4-vs-teensy-3.57283/ post #2). This ran fine for 12, 44.1, 48, 96, 100, 120, 144, 180, 192, 216 and 240 ksps. At 192 ksps, the base noise was roughly 8 dB higher at 192, not the 6 dB that goes with the increased FFT bin size. At 240 ksps the noise was about 10 dB above that at 192, so depending on the application, 192 may be the stopping point. At 384 it measured "something" but with noise all over the monitoring FFT and this is not useable.
I did a careful comparison at 48 and 192 ksps and the results are shown in the graph below. This test
included a sine wave introduced by a function generator. The major objective was to see the level of the
baseline noise, as reported above.
This graph shows the spectral power of the ADC output. Everything is the same for the two curves, except the sample rate. The 192 ksps rate curve should show the noise level as 6 dB higher due to the bandwidth of the FFT spectral bin being 4 times greater. It seems to be that, or slightly more. The weak harmonics of the sine wave are believed to come primarily from the function generator source, rather than the ADC. It needs an audio filter to be sure.
To get a measure of the noise spectrum and its variation, measurements of spectral power (FFT power) were
made at six different sample rates. The following graph shows the results. A reminder: these measurements were made
using the Teensy I16 library Quad I2S. This sets the input data to 16-bits. We will eventually revisit this
with 24-bit data available from the 1808 device.
More PCM1808 measured data is available on this page.
One caution is that Burr-Brown worked hard on the PCM5102 DAC to provide an output level of about 6 Volts, p-p, centered at about 0 Volts. This is great to drive external circuitry, but it is also able to damage the PCM1808 ADC. A 2:1 voltage divider is needed to protect that device along with a blocking capacitor to shift the DC level.
Issued 8 July, 2025. All Copyright © Robert Larkin 2025.
