Phase 2: Build the instrumentProjectA semester of eveningsAbout $180 for two assembled boardsTier 2

Project A: The ADS1299 board

Eight channels of 24-bit EEG around the chip inside most research and consumer systems. Schematic in KiCad, fabricated and assembled by JLCPCB, firmware on an ESP32, streaming into BrainFlow. A semester of evenings, and the whole signal chain becomes yours.

AssumesProject B: One channel of you KiCad to JLCPCB Firmware and SPI for a biopotential chipSpineAnalog / mixed-signal hardware

You are skimming: the title, the first figure, and the short version. Switch to Read in the header for the full page, or Deep to open every deep dive.

The Texas Instruments ADS1299 is a single chip that contains eight instrumentation amplifiers, eight 24-bit delta-sigma converters, a bias drive amplifier, lead-off detection, and a test-signal generator. It is inside the OpenBCI Cyton, inside several clinical amplifiers, and inside more than one consumer headset. Building a board around it is the audition for the hardware spine. You draw the schematic, lay out the board, have it made and assembled for the price of a textbook, write the firmware that configures it and streams samples, and end up with an instrument whose every design decision you can defend.

Predict before you look

The ADS1299's reference is 4.5 V and its maximum gain is 24. At that gain, how big is one step of the 24-bit converter?

About 22 nanovolts. Full scale is ±4.5 V divided by the gain of 24, so ±187.5 mV, or 375 mV total. Divide by 2²⁴ (about 16.8 million) and each step is 22.4 nV. The chip’s own noise is around a microvolt, so the last five bits or so are noise. That is normal for a delta-sigma converter and it is why “24-bit” is a description of the arithmetic, not the accuracy. The explainer on analog to digital is about this.

Parts

Part	Where	Qty	Approx.
ESP32 DevKit (or Raspberry Pi Pico) For streaming from the ADS1299 board over USB later. Either board works; pick one ecosystem and learn it.	Adafruit, SparkFun, Amazon	1	$8
Pinecil soldering iron (USB-C powered) and a spool of solder Or any temperature-controlled iron. Lead-free solder is fine; get flux.	Pine64 store	1	$35
USB isolator (ADuM3160/4160 based) Lets the board talk to a laptop on charger power without connecting the laptop's ground to you. Batteries are still simpler.	Amazon, Adafruit	1	$20
USB oscilloscope (Hantek 6022BE class) or a Rigol DHO800 if budget allows A real bench scope is better; the engineering building has them. A USB scope on your desk is what you will actually use at midnight.	Amazon, Rigol	1	$70
Your ADS1299 board, fabricated and assembled by JLCPCB (5 boards, 2 assembled) The ADS1299 itself is ~$60 in single quantity and dominates the cost. Order two assembled so one can die.	JLCPCB	1	$140
Total (prices drift; treat as a ceiling)			$273

The design, stage by stage

Start from TI’s evaluation board schematic and the OpenBCI Cyton schematic, both public. Do not invent; copy and understand, then change one thing at a time.

Power. A 5 V analog rail (AVDD) from a low-noise regulator and a 3.3 V digital rail (DVDD) for the chip’s logic and the microcontroller. If your microcontroller is a 3.3 V part, DVDD at 3.3 V means no level shifting. The Cyton runs the analog side at ±2.5 V instead of 0 to 5 V; either works. Every supply pin gets a 0.1 µF capacitor within a few millimetres, and the reference pins get the capacitors the datasheet specifies to the microfarad.

Inputs. Each channel is a differential pair. For EEG you usually tie all the negative inputs together to a single reference electrode through the chip’s SRB pin, so each channel measures its electrode against the common reference. A small series resistor (a few kilohms) and a capacitor on each input give input protection and some radio-frequency filtering. Connect through touch-proof (DIN 1.5 mm) sockets so that ordinary EEG electrodes plug straight in.

Bias drive. The chip’s BIAS amplifier is the driven-right-leg circuit you built by hand in Phase 1, now built in. Its output goes through the series resistor to a third electrode. The internal bias reference can be set to mid-supply.

Digital. SPI to the microcontroller: clock, data in, data out, chip select, plus the data-ready line that the chip pulls low each time a sample is available, and start and reset lines. Keep the digital traces away from the analog inputs.

Layout. Analog on one side, digital on the other, one ground plane split logically so digital return currents never cross under the analog inputs. Input traces short, paired, and symmetric. A guard ring around the input section. The datasheet’s layout section is short and every sentence in it is there because someone got it wrong.

Assembly. The ADS1299 is a 64-pin quad flat pack with 0.5 mm pitch. You can hand-solder it with flux and patience, but JLCPCB will place and solder it for a few dollars, and the chip is expensive enough that you want the reflow done by a machine. Order the passives from their parts library so everything can be assembled in one go.

Bring-up, in order

Before the chip is powered, check every supply rail with the multimeter with the microcontroller disconnected. 5.0 on AVDD, 3.3 on DVDD, nothing shorted.
Connect the microcontroller. Read the ID register over SPI. The ADS1299 returns a fixed value (0x3E) and until you read it, nothing else matters. If you cannot, check chip select polarity, SPI mode (the chip wants clock polarity 0, phase 1), and that the clock is running.
Turn on the internal test signal (a square wave of known amplitude) and route it to all channels. Read data frames on each data-ready pulse: 3 status bytes then 8 × 3 bytes of channel data, two’s complement. Convert to volts. You should see the square wave at exactly the datasheet’s amplitude on every channel. This validates the SPI, the framing, the sign handling, and the scaling in one step.
Short all inputs to the bias reference on the board and record sixty seconds at gain 24 and 250 samples per second. Compute the RMS in 0.5 to 40 Hz. This is your noise floor; a good board is around 1 µV RMS. Write it down. It is the first number on your characterization report.
Connect electrodes: eight on the head or two for a start, reference to SRB, bias to the forehead. Run the eyes-closed test. Alpha, on eight channels, from a board you designed.
Stream to the computer over Wi-Fi or Bluetooth using the OpenBCI Cyton’s serial protocol, so that BrainFlow and every tool that speaks to a Cyton speaks to your board too. Or write your own simple protocol and a small Python receiver. Then wrap it as a Lab Streaming Layer outlet.

What goes wrong

The ID register reads back 0xFF or 0x00. No SPI communication. Chip select is inverted, or the clock is too fast (start at 1 MHz), or the chip is not out of reset (hold RESET high, then send the SDATAC command before reading registers). Test signal looks right on channels 1 to 4 and wrong on 5 to 8. A byte-order or frame-alignment bug in the firmware; you are reading the frame one byte off. Enormous noise, hundreds of microvolts, on every channel. The reference buffer is off (check CONFIG3), or the bias drive is not connected to the body, or the board is being powered from a laptop. One channel is noisy, the rest are fine. A solder bridge or an open on that channel’s input network. Reflow it. Alpha appears and then the board resets every few seconds. The regulator’s dropout or a brown-out on the Wi-Fi transmit burst. Bigger bulk capacitance, or a better regulator.

The registers you will actually set

Start with the values below, then read the register map for each one until you can say what every bit does. Typical starting points for EEG at 250 samples per second: CONFIG1 sets the data rate; CONFIG2 controls the test signal; CONFIG3 turns on the internal reference buffer and the bias amplifier; each CHnSET sets that channel’s gain to 24, connects it to a normal electrode input, and connects its negative side to the SRB2 reference pin. The exact hexadecimal values are in the datasheet and in the OpenBCI firmware, and copying them without reading the map is how mysterious behaviour happens later.

Stop here if you like. Eight channels of your own alpha is the milestone. Log it, then go to the phantom head to find out what the board actually does.

Where this leads

The phantom head gives you known signals to measure your board against. The characterization report is where the numbers go. And the explainers on amplifiers, conversion, noise, and safety each explain a design decision you just made.

ArtifactEight channels of my own alpha from a board I designed, plus a measured noise floor in µV RMS over a stated bandwidth.

Recall

What is the first thing to do after powering an ADS1299 board, and why?

Read the ID register over SPI (expect 0x3E). Until that works, nothing else can be debugged; it validates power, SPI wiring, mode, and reset.

Recall

Why validate with the internal test signal before connecting electrodes?

It is a known-amplitude square wave routed to every channel, so it checks SPI framing, sign handling, byte order, and voltage scaling all at once with no biological uncertainty.

Recall

Why does 24 bits not mean 24 bits of accuracy here?

At gain 24 one step is about 22 nV but the chip's own noise is around 1 µV, so the last several bits are noise. Effective resolution is what the noise floor allows.