EMG Acquisition and Calibration

Electrodes and differential acquisition

Two surface electrodes measure a voltage difference associated with the combined activity of muscle fibres beneath them. An instrumentation amplifier increases that differential signal while rejecting the voltage common to the two inputs. The useful operating range depends on electrode offsets, common-mode voltage, gain, and supply headroom.

We used AD620-based acquisition in the early MuscleMate system and moved toward integrated multichannel biopotential conversion as the hardware developed. Analog filtering, gain control, and input-range management prepared the signal for digitization. Those choices also influenced the features available to the recognizer, since clipping or a change in frequency response changes the measured spectrum.

Textile sensing added a mechanical dimension. Electrode pressure, contact area, and movement affected the electrical interface. The fitted band and conductive connections therefore had to be designed together with the acquisition electronics.

We developed both laminated textile assemblies and knitted conductive connections. This brought several electrical questions into the garment itself: how to keep a sensing pad in repeatable contact with the skin, how to insulate its conductive route, and how to connect that route to removable electronics. Conductive-yarn choice, electrode attachment, band tension, and grip all influenced the resulting interface.

The sensing pad and its lead served different purposes. The pad established the intended contact area; insulation along the lead confined skin contact to that area. The same distinction applied on the outside of the garment, where the conductive route required protection from the surrounding environment.

Calibration during exercise

We characterized skin impedance while recording EMG during cycling. The setup used a 50 kHz impedance measurement, selectable feedback and calibration resistors, and an AD5933 evaluation system. At the same time, the wearable streamed EMG over Bluetooth to a tablet.

Trials lasted 10–30 minutes, with impedance readings every 30 seconds. We observed impedance decreasing after perspiration began, over a timescale of minutes. That behavior matters because a reference recorded immediately after fitting the band can represent a different contact condition from the one present later in the session.

The calibration problem has two parts. Interface characterization concerns the transfer from the skin to the electronics. Reference-contraction normalization sets the scale used to interpret muscle amplitude. Treating these separately gives each correction a clear physical role. The physiological-acquisition patent describes the use of impedance-derived calibration data in EMG processing.

Multichannel electronics

The multichannel development used an ADS1298 acquisition front end, connected to the processor over SPI. Conversion-ready signaling allowed the firmware to coordinate data reads with the arrival of samples. The motion sensor used I²C, flash memory used SPIFI, and the radio had its own serial interface.

Integrating these peripherals required explicit attention to the timing and state of each stream. Biopotential data arrived at a much higher rate than the motion measurements, while the radio and flash introduced their own transfer schedules. Buffers decoupled acquisition from processing and communication.

A 24-bit channel sampled at 8 kHz generates 192 kbit/s, or 24 kB/s, before packet framing. This made the division between raw-data recording and on-device feature computation a practical architecture decision. Local storage supported detailed recordings, while embedded algorithms supplied compact results for real-time feedback.

Multiplexed biopotential acquisition

We implemented a multiplexed acquisition arrangement that connected twelve input paths to four 24-bit converter channels. A switching network and embedded control expanded the set of electrode signals that the electronics could acquire within a compact assembly.

The scheduling problem is as important as the channel count. Each selection needs time for the analog path to settle, conversion to complete, and the samples to be assigned to the correct input. With twelve inputs sharing four channels, an evenly divided schedule has three selections per complete scan. The effective rate for each input follows from that scan period.

The prototype integration combined the analog front end, accelerometer, flash memory, and Bluetooth Low Energy link, including communication with a phone. Board development reached a 37 mm diameter at this stage. Bringing these elements together required acquisition timing, local buffering, and communication to operate as one system.

Firmware therefore participates directly in the measurement: it controls the switches, coordinates converter reads, and retains the association between samples and electrodes. These timing details also matter when comparing activation between muscles or aligning EMG with acceleration.

Bringing the hardware into operation

We brought up the electronics in stages, beginning with unpowered resistance checks and then verifying the 5 V USB input, 3.7–4.2 V battery input, and regulated 3.3 V supply. Clock and processor checks followed power sequencing. A 12 MHz system clock and ISP programming established the basis for testing the peripheral interfaces.

The acquisition tests exercised SPI clock and data lines together with the converter’s DRDY signal. Separate tests checked acceleration, flash memory, and the radio before the full application firmware combined them. This sequence made it possible to distinguish an electrical or interface problem from an algorithm operating on the resulting samples.