Subcutaneous Transmitter (A3049)

Contents

Description

[05-JUL-23] The Subcutaneous Transmitter (A3049) is an implantable telemetry sensor for mice and rats that provides amplification and filtering for up to two independent biopotentials. When equipped with a small coin cell, it is fits comfortably in a mouse. When equipped with a large coin cell, it fits comfortably in a rat. The A3049 operates with our Subcutaneous Transmitter system.

Figure: Subcutaneous Transmitter A3049A. This specimin is equipped with 200-mm leads and A-Coil terminations on each lead.

The A3049 amplifier can provide gain of ×100 for frequencies up to 620 Hz. The logic may be programmed to sample at 64, 128, 256, 512, 1024, or 2048 SPS. The low-pass filter may be configured for a corner frequency of 20 Hz, 40 Hz, 80 Hz, 160 Hz, 320 Hz, or 640 Hz. The input high-pass filter provides a corner frequency of 0.16 Hz, but may be removed to give gain all the way down to 0.0 Hz. All versions of the A3049 are equipped with 0.7-mm diameter red and blue leads, and a clear-coated loop antenna. The length of the leads, the battery loaded next to the circuit, the operating life, the termination of the leads, the sample rate, the gain of the amplifier, and the bandwidth of the amplifier all vary from one version to the next.

Figure: Subcutaneous Transmitter A3049E. This specimin is equipped with 130-mm leads and A-Coil terminations on each lead.

The A3049 may be configured as a single or dual-input sensor. As a single-input sensor it will transmit one signal on one telemetry channel. As a two-input sensor it will transmit two signals on two telemetry channels. The first channel number will always be an odd-numbered channel. The leads loaded on the transmitter depend upon its input configuration, as shown below.

Each transmitter has a label providing two numbers. The first is a batch number, B, the second is a telemetry channel number, N. A single-channel transmitter uses channel N only. A dual-channel transmitter uses channel N for the X input and N+1 for the Y input. In a dual-channel transmitter, N is always odd.

The specification of the transmitters depends upon their version, but we give an example specification in the table above.

Versions

[26-MAY-23] The Subcutaneous Transmitter (A3049) can be equipped with a dozen different batteries, any lead length up to 280 mm, several lead diameters, one or two recording channels, a dozen varieties of lead terminations, several types of antenna, and a range of bandwidths, gains, and sample rates. You specify which transmitter you want with a full SCT part number. The part number begins with A2049 and is followed by the primary version letter that tells us the battery we load on the circuit, and the input type (I, II, III, or IV as shown above. Following the letter we have one or two more numbers and letters that specify the sample rate of the inputs. We use the numbers 1-5 to indicate 128, 256, 512, 1014, and 2048 SPS respectively. We use the letter "Z" to indicate that the low end of the frequency response reaches all the way down to 0.0 Hz. After a dash we have a number and letter to specify the length and type of the leads. After a second dash we have letters specifying the electrodes, and after a final dash we have a letter specifying the antenna.

Figure: A2049 Part Numbering Scheme. Click on the large boxes to jump to tables listing letter codes and options.

The table below lists the A3028 primary version codes. Battery capacities are usually expressed in units of mA-hr (milliamp-hours). We convert to μA-dy (microamp-days) so it is easier to divide the capacity by the active current and obtain the operating life in days. We give frequency response in Hertz, sample rate in samples per second, and dynamic range of each input in millivolts. Input resistance is either 10 MΩ or 20 MΩ, see Amplifiers.

For each analog input we specify the bandwidth, sample rate, input dynamic range in millivolts, and channel number offset. In terms of ADC counts, the dynamic range is always 0-65535, as produced by a sixteen-bit ADC. The zero-value of an input is the sample we obtain when we short the two inputs together. The zero value depends upon the battery voltage, VBAT, according to Z = 1.8 V × 65535 ÷ VBAT. The dynamic range is the battery voltage divided by the gain of the amplifier. When we specify dynamic range, we assume VBAT = 2.7 V, which is true for most of the life of a lithium primary call. When the amplifier gain is 100, the dynamic range is 27 mV.

The shelf life of a transmitter is how long it takes to exhaust the battery permanently when we leave it inactive on the shelf. See below for details of current consumption and how to calculate battery life of new versions of the A3049. By default, we set the top of the frequency range at one third the sample rate. The A3049's low-pass filters provide 20 dB of attenuation at one half the sample rate. Frequencies above one half the sample rate will be distorted by sampling, and so compromise the fidelity of the recording. Because the EEG signal contains less and less power as frequency increases, this attenuation is sufficient to ensure that distortion is insignificant.

Leads

[19-APR-23] We define the lead names and provide links to photographs and drawings of the leads in the A3028 Manual's Leads section. We present the various antennas we have used for implants in the Antenna section. The best antenna for rat implantation is the 50-mm A-Antenna.

Electrodes

[19-APR-23] We define the electrode names and provide links to photographs and drawings of the electrodes in the A3028 Manual's Electrodes section.

Analog Inputs

[18-JUL-23] The A3049 provides up to four signal inputs: X+, X−, Y+, and Y−. Each of these inputs has a reserved color for its leads: red, blue, yellow, and green respectively. These four leads are present or absent in accordance with each transmitter's lead configuration. Whenever the X+ (red) lead is present, it uses the X− (blue) lead as its reference potential. When the Y+ (yellow) lead is present without the Y− (green) lead, the Y+ lead uses the X− lead as its reference potential. When the Y− lead is present, the Y+ uses the Y− lead as its reference potential. When equipped with three leads, the A3049 is a two-channel sensor with a shared reference potential. When equipped with four leads, it is a two-channel sensor with separate reference potentials.

Figure: Response of Batch of A3049B3 to 10-MΩ Sweep.

The impedance of X input, as seen at the tips of its electrode leads, is 10 MΩ. When the Y input uses X− as its reference, the Y input impedance is 10 MΩ. When the Y input uses Y− as its reference, the Y input impedance is 20 MΩ. Most transmitters provide a high-pass filter by placing a capacitor in series with the input. The corner frequency of this high-pass filter is 0.2 Hz. When the input impedance is 10 MΩ, the high-pass filter presents a 100-nF capacitor in series with the input, and when the input impedance is 20 MΩ, the series capacitance is 50 nF. When we modify the transmitter to remove the high-pass filter, these capacitors will not be present at the input.

The X signal is supposed to be a measure only of difference between X+ and X−. The average voltage of X+ and X− is the common mode voltage on X, and the difference between X+ and X− is the differential mode voltage. Suppose we apply the same sinusoidal voltage to both X+ and X−. The common mode voltage is the sinusoidal voltage and the differential mode voltage is zero. Under these circumstances, we would like X to be zero, but instead we will see a trace of the common-mode voltage appearing in the X signal. The ratio of the common-mode voltage amplitude and the X signal amplitude is the common mode rejection ratio, or CMRR. The plot below shows how the CMRR of X and Y vary with frequency.

Figure: Common Mode Rejection Ratio (dB) of X and Y for the A3049A3. We apply a 15-mVpp 100-kΩ common-mode sweep to each input while measuring the recorded signal amplitude.

The X-input provides CMRR of 40 dB for frequencies for frequencies below 160 Hz. The signal we see on X will be 1% the amplitude of the common-mode signal we apply to X. The CMRR of the Y-input is >40 dB for frequencies below 10 Hz, but drops for higher frequencies.

The distortion of a signal by our telemetry system is the extent to which it changes the shape of a signal. We apply a 10 mVpp sinusoid to the X and Y inputs of an A3049AV3. The AV3 is equipped with two 160-Hz amplifiers. Input dynamic range is 27 mV. We increase the frequency from 1/8 Hz to 200 Hz. For each frequency, we obtain the spectrum of the signal and measure the power outside the sinusoidal frequency as a fraction of the sinusoidal power using this script. We express the result in parts per million.

Figure: Distortion of 10-mVpp Sinusoid versus Sinusoidal Frequency. Non-sinusoidal power as a fraction of sinusoidal power in parts per million. Sine wave generated by BK Precision 4053B, specified total harmonic distortion <1 ppm.

The distortion of the X is dominated by random electronic noise. There are no significant peaks in the spectrum outside the fundamenta.

Figure: Spectrum with 50-Hz, 10-mVpp Sinusoid. Horizonal: 10 Hz/div. Vertical: 0.4 μV/div. The peak is 4000 μV.

We note that the distortion generated by the A3049 is hundreds of time less powerful than that of its predecessor, the A3028. The A3048 samples the signal uniformly, thus eliminating the scatter noise present in the A3028 signal.

Design

[25-APR-23] Details of the design are available in the following library of design files. Note that all our designs are protected by the GNU General Public Lisence.

[20-JUL-23] The following table lists versions of the assembled A3049 electronic circuit, out of which we make the A3049-series transmitters.

Modifications

[08-SEP-23] Here we list the electronic circuits we can use to assemble the various types of A3049 transmitter, and the modifications required by that circuit prior to assembly.

Synchronization

[19-APR-23] When we want to mark in our SCT recordings the time at which some event took place, such as the start of a video recording, the moment that a light was flashed, or when an noise commenced, we can use an auxiliary SCT to record a synchronizing signal along with the signals received from implanted SCTs. See the Synchronization section of the A3028 manual for details.

Body Capacitance

[19-APR-23] See Body Capacitance in the A3019 manual.

Battery Life

[19-SEP-23] We equip all our subcutaneous transmitters with CR-series lithium primary cells. The voltage produced by these batteries begins at around 3.0, drops to 2.8 V for most of the battery's life, and drops rapidly towards the end of life, as shown below for CR1025 batteries.

Figure: Discharge of CR1025 Batteries. Discharge current is 75 μA, battery capacity is nominally 30 mAhr.

To obtain the minimum operating life of an A3049 transmitter, we divide the battery capacity in μA-days by the maximum current consumption in μA, and then subtract one day. The subtraction of one day is necessary to account for the twenty-four hours of testing we perform on each transmitter during quality control. To obtain the maximum current consumption of an A3049 transmitter, we use the following relation.

In the above relation, we have 22 μA base current consumption, which powers the logic chip (15 μA), amplifiers (4 μA), and miscellaneous circuits (2 μA). Additional current consumption by digitization and transmission is 0.11 μA per sample per second, or we could say that each sample requires 0.11 μC of charge drawn from the battery. The above formula predicts 50 μA for 256 SPS. The average current consumption of the A3049 circuits is roughly 10% lower than the maximum.

Figure: Current Consumption versus Total Sample Rate. Straight line fit slope 0.095 μA/SPS, intercept 20.0 μA.

In the table below, we use our formula for maximum current consumption and combine it with the nominal capacity of the batteries we might use with the A3049. The CR1620 is the smallest battery we believe we can load onto the 20-mm diameter circuit. The CR2477 is the largest battery we know for sure that a large adult rat can tolerate.

Figure: Minimum Operating Life in Days for Various Batteries.

In each of the above entries, we have divided the nominal capacity of the battery by the maximum current consumption and subtracted one from the result to obtain our minimum operating life.

Encapsulation

[19-APR-23] All versions of the A3049 are encapsulated in black epoxy with a coating of silicone. The silicone is "unrestricted medical grade", meaning it is approved for implants of unlimited duration in any animal, humans included.

Development

[19-APR-23] Start circuit design. Enhance the second stage of the amplifier so it can provide gain greater than 2.5. We are looking at this circuit: an op-amp with a positive feedback capacitor, which provides two poles for a filter function.

Figure: Two-Stage Low-Pass Filter. We use this circuit to create the second and third poles of our low-pass filter.

When Ra = Rb, and the capacitors are equal, we must set the gain of the circuit to 2.5 to obtain the second and third poles of our desired 0.5-dB Ripple Chebyshev response. We want to use capacitors of the same value, but we are prepared to use any resistor values. We want to decrease the gain in the first stage, so we can use op-amps with lower bandwidth. We want to increase the gain of the second stage so we can use op-amps that are stable only for gain ≥5, such as the MAX4474. We consider letting the first resistor, Ra, on the left be increased by a factor of α and the second, Rb, be increase by β. To obtain the correct response, we find that αβ = 1 and α is a function of the gain, as shown below.

Figure: Amplifier Second-Stage Gain and Resistor Values. We use a for α, b for β, and A for the gain. We try values for Ra and see what Rb has to be, then calculate actual gain if we were to use Ra and Rb for R2 and R1 respectively.

If we set the gain to 5, we find that we can use 500 kΩ for Ra and 2.0 MΩ for Rb. Furthermore, we can use 2.0 MΩ for R1 and 500 kΩ for R2 and obtain gain of exactly 5.

[23-MAY-23] We receive twenty of A3049AV1. These circuits use the MAX4474 40-kHz amplifier for X and Y. Program with P3049A01. Apply 3-Vpp sweep. In each sweep, we have one of X or Y connected to the sweep through a resistance of 10 MΩ or 100-kΩ, and the other channel open circuit.

Figure: Sweep Response of Unmodified A3049AV1 programmed as A3 (Dual-Channel 512 SPS).

The input resistance of Y+/Y− is 20 MΩ, while that of X+/X− is 10 MΩ. The MAX4474 op-amp we use for gain in both amplifiers has a gain-bandwidth product of 40 kHz. The first amplifier stage has gain 20. We predict the bandwidth of the unfiltered amplifier will be 2 kHz. We measure current consumption of three circuits from 64 SPS to 2048 SPS. Intercept is 18 μA + 0.097 μA/SPS.

[26-MAY-23] We vary sample rate and measure active current, see here.

[01-JUN-23] Nathan is applying sparks to the antenna of an A3049AV1 and loading various parts in its antenna matching network (L1, C17 and C6 in S3049A_1.gif) to see if we can reduce the probability of U9, the VCO, being damaged by static electricity. Our spark source is a plasma ball, see here. A direct spark to U9-7 always destroys the VCO. A spark to the antenna with C6 = L1 = 1.0 nF, C17 = OMIT (Network A) usually destroys the VCO, but not always. With C6 = 27 nH, C17 = 200 Ω and L1 = 1.0 nF (Network B), no amount of plasma ball sparking will destroy the VCO, but U9-7 is now connected to VB, which we long ago determined to be a sign of impending failure. After three days running, however, this allegedly damaged transmitter is still running.

With C6 = L1 = 15 pF and C17 = 200 Ω (Network C), no amount of plasma sparking will damage the VCO. The transmitter keeps running, current is normal, and U9-7 has no DC voltage. Nathan sets up the plasma ball and records sparks with a 100-MHz, 2 GSPS oscilloscope equipped with a 500-MHz ×10 probe. We have Network C in place.

Figure: Typical Plasma Ball Spark, Network C. Blue: Voltage on antenna when we touch it to the spark source, offset −27 V. Yellow: VCO output, U9-7. Green: ground reference on circuit board.

The sparks have rapid rising edges, but the peak voltage is attained only after several such steps. In the trace belowl, we see the spark enduring for around 12 μs and reaching 100 V.

Figure: Typical Plasma Ball Spark. Blue: Voltage on antenna when we touch it to the spark source, offset −27 V. Yellow: VCO output, U9-7. Green: ground reference on circuit board.

Network C is a high-pass filter with corner frequency 50 MHz, so it rejects everything except the jumps in the spark voltage. We compare radiated power with Network C and Network A. We see no significant difference between them. Perhaps Network C is 0.6 dB less powerful than Network A.

[05-JUN-23] We try L1 = 0Ω, C17 = 200Ω, C6 = 15pF, a high-pass filter looking from antenna to VCO with corner frequency 50 MHz. Nathan reports, "This matching network gives a normal power output, but after one shock the transmitter begins consuming 11 mA. This is due to a single shock being able to short the VCO output to VB, allowing for a DC current to flow to ground. This matching network is NOT static proof." We try L1 = 10 Ω, C17 = 27 nH, C6 = 0Ω, a high-pass filter in which the 27 nH interacts with the antenna source impedance. Nathan reports, "After one shock from the plasma ball on the antenna of this transmitter, the transmitter ceases to function. Its current consumption increases to 18mA, indicating to me that the one shock to its antenna permanently damaged the VCO." We conclude that a two-component network that does not cause significant transmit power loss cannot protect the VCO from spark damage. We will use Network C in the A3049BV1.

[08-JUN-23] We receive 80 of A3049AV2 with two 160-Hz amplifiers. Test first circuit, all components loaded correctly. The X-input behaves perfectly. The Y-input has correct gain and frequency response, but is offset upwards. Average value of X for VB = 2.70 V is 43800 (from which we would deduce VB = 2.69 V) and for Y is 56400. We measure with our DVM VC = 0.0 V, U6-4 = −0.2 mV, U5-2 = 4.7 mV, U5-3 = 0.2 mV. Our U5 appears to have an input offset voltage of +4.7 mV, which when multiplied by ×100 gives us a final offset of 470 mV. We also notice a bug in the firmware: the X-channel has the higher of the two channel numbers. We try an A3049AV1 for which X is 47000 and get U6-4 = −03.3 mV, U5-2 = 4.6 mV, U5-3 = 0.2 mV.

[13-JUN-23] We load OPA2369 in place of MAX4471 for U5. The OPA2369 has a maximum input offset voltage of 0.25 mV, compared to 7 mV for the MAX4471. The Y input average drops from 56400 to 40291, compared to 40300 for the X input.

[14-JUN-23] We compare OPA2369 (12 kHz), MAX4471 (9 kHz), and MAX4464 (40 kHz) amplifier gain with no capacitors. The MAX4464 X-channel amplifier output is unstable with no capacitors. With 10-MΩ sweep connected, the output is noisy, but we are able to measure amplitude. The MAX4464 is operating with gain ×5 in the second stage, which is the minimum gain for which it is stable. The other two op-amps produce a stable output.

Figure: Amplifier Frequency Response for Various Op-Amps.

The bandwidth is limited by the ×20 gain of the first stage. We expect bandwidth 12 kHz ÷ 20 = 600 Hz for the OPA2369 and 9 kHz ÷ 20 = 450 Hz for the MAX4471. We observe 3-dB bandwidth 400 Hz for both. We must load the OPA2369 for the Y-channel to avoid large DC offsets in the amplifier output. Our Y-amplifier will support SCT bandwidths up to 320 Hz at 1024 SPS. With MAX4464 loaded in the X-amplifier, we expect bandwidth 2 kHz, permitting sample rate 4096 SPS. Our A3049AV3 and AV4 circuits will support dual-channel up to 1024 SPS or single-channel up to 4096 SPS. If we want faster dual-channel transmitters, we must revert to the OPA2349 70-kHz amplifier, of which we have one thousand on the shelf.

[15-JUN-23] We load OPA2349 for U5, replacing an OPA2369, and find that the Y-amplifier output rises from 44 kcount, which is the same value we have for X, to 48 kcount because of the OPA2349 input offset voltages. We measure sweep with 240 pF capacitors loaded in the amplifier. We do the same for MAX4474 and MAX4471. The MAX4474 is stable with 240 pF. The MAX4471 provides greater bandwidth with 240 pF.

[27-JUN-23] We have twenty-three A3049A3-AAA-B200-B made with the A3049AV2. The long leads on these devices pick up lots of mains hum and other noise. The Y-channel offsets generated by the MAX4474 are excessive. We provide 2.64 V to each device in turn, place lead tips in water to reduce mains hum, and measure the average value of X and Y, then calculate the equivalent offset voltage across U5-2 and U5-3. Half of the circuits we could allow to proceed without modification, but the other half require that replace U5 with OOPA2369. We resolve to replace U5 on all of them.

Figure: Offset of Y from X with MAX4474 for U5.

When both amplifiers are saturated with mains hum, current consumption of the A3049AV2 programmed as 49A3 rises from 125 μA to 155 μA. With only one saturated with mains hum, increase is 15 μA. When we perform the same test on an A3028KV2, we see increase an increas of only 5 μA.

[28-JUN-23] We replace U5 on two of our A3049A3-AAA-B200-B. With a battery as power supply, X and Y have the same average value within a ±500 μV. But when we connect an external power supply, we see the Y input of every circuit dropping down, sometimes saturating at zero, sometimes with a constant negative offset with respect to X. The X input remains centered where we expect it to be. We connect probes to Y+, X+, and Y−, grounding them to VC, and see the following.

Figure: Input Noise with Benchtop Power Supply. We have three probes grounded to VC. Yellow: Y+, Blue: X+, Green: Y−. Vertical 50 mV/div. Horizontal 2.5 μs/div.

Here we see the effect of the long leads: Y− has far less noise than Y+. The noise is around 300 kHz, far above the 12-kHz bandwidth of the OPA369 that buffers Y−, and of amplitude 100 mVpp. With our probes disconnected, current coming from bench-top power supply, and the three long leads loaded on X+, X−, and Y+, we see Y offset by −14 kcnt (−5.6 mV) with respect to X. We remove the Y− lead and now Y is offset by +0.6 kcnt (+0.2 mV). Reconnect Y+ and power with battery, Y offset from X by −0.4 kcnt (−160 μV).

We replace U5 on all 23 of our A3049A3-AAA-B200-B. Half of them have X and Y within 500 cnt. But the other half have extreme and fluctuating baseline value of Y. As we change the noise or mains hum on the inputs, the baseline of Y can shift. The baseline of X never shifts. We examine U6-1 and U6-3. We sometimes see pulses at 1 kHz on U6-3. We sometime see a −20 mV offset on U6-1. We replace all R21 = 10MΩ with R21 = 100 kΩ. We wash and dry. None of the circuits show noise on U6-3 nor offset on U6-2. Every Y is within 500 cnt of X. Noise on both inputs is sinisoidal mains hum of equal amplitude and matching phase. Update S3049A_1 so that R21 = 100 kΩ, U5 is OPA2369, C6 = L1 = 15 pF, and C17 is 200 Ω.

[30-JUN-23] With latest measurements, current consumption of the A3049AV2 has slope 0.095 μA/SPS, intercept 20.0 μA. Compare to latest A3048AV1 measured slope 0.106 μA/SPS, intercept 16.1 μA.

[03-JUL-23] We receive 99 of A3049AV3 and 105 of A3049AV4. We take three of each type to test. These boards have MAX4474 for U4 and OPA2369 for U5. In place of L1 and C6 they have 15 pF, and for C17 they have 200 Ω. The AV3 has 160-Hz corner frequency, the AV4 has 80-Hz corner frequecy. Each board is labelled with an index and version. We program as A3049AV3 and check frequency response, power output, and channel offset.

Figure: Offset of Y from X with OPA2369 for U5 in AV3 and AV4. We give the offset in mV with respect to the input dynamic range of 27 mV.

When we first program 180AV4, the Y channel offset is 5.9 mV. We see U6-3 at −10 mV and U6-1 at −8 mV. Circuit 066AV3 has Y offset 2.6 mV. We load R21 with 1 MΩ on both these boards and see the offsets drop to 0.7 and 0.9 mV respectively. When we load R21 with 100 kΩ, the offsets drop further to 0.12 mV and 0.48 mV, which is consistent with the ±0.75 mV input offset voltage of the OPA369. We resolve to replace R21 with 100 kΩ for all dual-channel transmitters that use VC as a common reference. When we use Y− as the reference for the Y-channel, we will replace R21 with 1 MΩ and see what happens.

[05-JUL-23] We have a batch of 23 of A3049A3-AAA-B200-B encapsulated. Frequency response and X-Y offsets are all fine, but noise in the Y channels is 20-30 μV consisting of single-sample spikes of up to 80 μV. The X-input noise is 10 μV. We trace this to a firmware timing error, combined with the Y-amplifier op-amp OPA2369 being slower than the X-amplifier op-amp MAX4474. We need to give the amplifiers time to drive the ADC input in advance of the next sample. We fix the problem by increasing the minimum number of clock ticks between selecting a new channel and the ADC conversion. We increase the End Clock Offset from 1 to 7 and measure noise on X and Y. We measure crosstalk also, by applying a differential mode signal to the other input of amplitude 15 mVpp, and measuring the ratio of the the test input amplitude to the other input's amplitude.

Figure: Increasing End Clock Offset in Firmware, Its Effect upon Noise and Crosstalk.

We enhance the firmware so it applies an offset chosen to suit the sample period. For 4096 SPS, the offset is only 3, but for lower sample rates, we increase the offset until for total sample rate 512 SPS and lower the offset is 15. In a batch of A3049H2, noise on X and Y is around 20 cnt.

[06-JUL-23] We re-program all 28 of A3049H2 with our new firmware. Total sample rate is 512 SPS, so we have ECK offset 15. After re-programming, noise on X and Y with leads in water settles down to around 12 cnt rms, or 5 μVrms. We study sampling noise with an A3049AV3.

Figure: ADC Readout. Yellow: RF Frequency HI. Blue: VA, analog power, AC-coupled, 20 mV/div. Green: ACTIVE. Red: CONVST. Horizontal 500 ns/div.

We drive CONVST LO while we read out the ADS8860. We drive it HI to initiate conversion. In our latest firmware, we keep CONVST LO until we are done with RF transmission. Driving it HI as soon as all ADC readout is done, before the RF transmission is complete, works just as well.

Figure: VA During Sampling. Yellow: RF Frequency HI. Blue: VA, analog power, AC-coupled, 20 mV/div. Green: ACTIVE. Red: CONVST. Horizontal 500 μs/div.

In the longer view, we see 20-mV sampling ripple on the analog power supply, VA. Right now we are using ECK Offset to give us time for the X and Y amplifiers to settle after selection with our analog switch. We try switching immediately after conversion, with the falling edge of ACTIVE, or the falling edge of Delayed ACTIVE, or the falling edge of Delayed Delayed ACTIVE. But neither offers lower noise.

Figure: Noise with Various Selection Criteria. We have total sample rate, EO = end clock offset, DA = delayed active, DDA = delayed DA.

[07-JUL-23] We are trying to understand why it is that switching between X and Y with U7 at !ACTIVE results in noise on X and Y, while switching at ECK does not. We synchronize CONVST with TCK, so that it goes low at the start of serial readout and goes hi after the last RF bit has been transmitted. We start with switching on ECK, inputs open-circuit. We switch on !ACTIVE, inputs open-circuit and see the spectrum of "antenna noise", which is the noise we get from a transmitter in water with a broken ground lead. We short the X and Y inputs to ground and measure the spectrum again.

Figure: Noise Spectrum with Various Selection Criteria. Left: Switching on ECK, X and Y open circuit. Left-Center: Switching on !ACTIVE, X and Y open circuit. Right-Center: Switching on !ACTIVE, X and Y shorted to ground. Right: Switching on !Active, transmit scatter suppressed, X and Y open circuit.

Instead of !ACTIVE, we try !DDA, which is a rising edge 60 us after !ACTIVE. We see the same open-circuit noise. We suppress transmission scatter, switch on !ACTIVE and the antenna noise vanishes. We restore transmit scatter, and add another CONVST low pulse two CK periods before ECK, so we are now converting twice during each sample period: once immediately after readout, another time before ECK. Antenna noise vanishes. Current consumption for this 1024 SPS device increases from 121.6 μA to 126.2 μA when we add the extra sample. We shorten the secondary CONVST pulse with the help of !CK, but this does not reduce the current consumption. We return to just one CONVST pulse, but we set it equal to !ACTIVE. Current is 121.3 μA. We examine the output of the switch, which is U7-5. We trigger on SEL, which goes HI to select X, LO to select Y. We look at CONVST as well. We are running at 512 SPS per channel.

Figure: Sampling Charge Injection. Yellow: SEL, switch select, 1 V/div. Blue: Switch Analog Output, U7-5, connects to ADC input AINP, AC-coupled 20 mV/div. Green: CONVST, ADC conversion occurs on rising edge, 1 V/div. Horizontal 100 μs/div.

We see AINP jump down upon selection of X, recovering with two bounces in 100 μs. Another negative bounce during ADC sample and hold, recovering in 100 μs. This X-signal is provided by the 40-kHz MAX4474. We see AINP jump down again upon selection of Y, but this time the recovery consists of only one bounce and takes 200 μs. The same thing happens after conversion. The Y-signal is provided by the 12-kHz OPA2369. We want to allow sufficient time after switching before conversion, which appears to be around 200 μs, or seven CK ticks. One way to do that would be to switch 30 μs after conversion. If we switch synchronously with conversion, the switching is scattered, and we see noise induced on our open-circuit inputs. We look at the path of SEL from its source on U11 to its destination on U7 and we note that it passes beneath both amplifiers, threading its way between resistors in the top layer. We speculate that the edges of this signal induce noise in the amplifier input, but we note that SEL does not pass close to the amplifier inputs XP and YP. When we short the amplifier inputs, this SEL-induced noise disappears.

[18-JUL-23] Measure distortion in the X and Y inputs of the AV3. The distortion is dominated by transmission scatter for frequencies above 10 Hz.

[19-JUL-23] Eliminate scatter noise in firmware P3049A05 by adding a pulse on CSS during the last CK period during the sample period. We synchronize CSS with TCK during the serial readout, and this readout pulse remains shorter than the ACTIVE pulse. We now have two conversions per sample period and one readout. Current consumption of the A3049A3, with 2 × 512 SPS rises from 116.7 μA to 118.9 μA, or 0.002 μA/SPS, which is negligible compared to our existing 0.11 μA/SPS. Total harmonic distortion now remains below 20 ppm and is caused only by random noise.

[01-AUG-23] Discover bug in P3049A05, whereby for versions 21 and 31 and base id 87 we get x_id 87 and y_id 216, due to top bit of ID being inverted for second channel. We conclude this is an ABEL compiler bug. We work around the bug by calculating the completion code, cc, directly from x_id and y_id in our sample counter state machine, rather than from the intermediate id nodes, which are eliminated later by the compiler anyway.

We have encountered five A3049AV3/4 assemblies with U1 tilted on the board and one of the pins not soldered. The picture below is an example. We reflow the joints and the component sits properly on its footprint, circuit works.

Figure: Tilted U1 with Bad Joint.

[03-AUG-23] We have our first batch of A3049K1 with wires loaded, ready for QC1. The K1 is a dual-channel transmitter with 128 SPS and 0.3-40 Hz on X for EEG and 64 SPS and 0.3-80 Hz on Y for EMG. Current consumption is 38-42 μA. Noise is 5 μV on X and 6 μV on Y.

[10-AUG-23] We drain three of each of four types of battery over the course of a week: CR1025, CR1225, CR2330, and CR2450. We drain each battery with a resistor. Twice a day, we measure the battery voltage, disconnect the resistor, and measure the battery voltage again. For each type of battery we use a fixed resistor. We model the battery as a voltage source in series with a source resistance. We calculate this source resistance from our open-circuit and loaded battery voltage measurements, and our knowledge of the load resistance. We define the nominal drain current as the nominal battery voltage divided by the drain resistor. We take the nominal battery voltage to be 2.7 V. We plot the source resistance versus charge drained from the battery.

Figure: Effective Source Resistance versus Charge Delivered for Various Batteries. Click for higher resolution. We give nominal capacity from data sheet and our experiment's nominal drain current for each battery type. Plots by Calvin Dahlberg. Measurements by Calvin Dahlberg and Raphael Beauchemin.

The capacity of these batteries decreases with drain current. The CR1025 capacity of 30 mAhr, for example, is specified for a drain current of 0.05 mA. The capacity decreases as drain current increases. Our CR2450 drain resistor was 630 Ω. The CR2450 data sheet specifies capacity only for loads 1 kΩ and higher. The plot below implies that the capacity for our 630 Ω load could be as low as 300 mAhr, and indeed this is what we see in our drain plots.

Figure: CR2450 Capacity versus Load Resistance. Taken from the CR2450 data sheet.

We are draining all our batteries at higher than their maximum recommended rate. The effective source resistance remains roughly constant during the drain. One of the features of our A3048 and A3049 transmitters is the 2 mA burst of current they need when we turn them on. We conclude from the above study that, if a battery can power up an A3048 or A3049 when it is fresh, then it can do so for its entire operating life.