Subcutaneous Transmitter (A3019)

Contents

Description

The Subcutaneous Transmitter (A3019) is an implantable telemetric sensor for monitoring EEG. It succeeds the Subcutaneous Transmitter (A3013), being a smaller implementation of an almost identical circuit. See Subcutaneous Transmitter System for an overview of the transmitters and data-acquisition.

Figure: Subcutaneous Transmitter A3019A with Screw Electrodes. This transmitter is designed for use in mice. Volume is 1.0 ml, width 12.5 mm. Operating life is 4 wks at 512 SPS. The antenna is stainless stee cable coated with silicone for maximum flexibility in a mouse. The helical, stainless steel leads are insulated with five thin coats of silicone. The X+ lead is coated white at the end. The screws are M0.5, 1-mm long stainless steel with binding heads.

The A3019 comes with two helical steel leads coated with silicone. These detect biopotentials such as EEG or EKG. The transmitter is also equipped with a 50-mm stranded steel silicone-insulated antenna. The standard version of all A3019 transmitters comes with stainless steel screws soldered to the ends of the leads. The screws act as the electrodes when inserted into a hole drilled in an animal skull. We discuss the various options for electrodes on the tips of the leads in Electrodes.

The A3019D is the same as the A3019A, but with a larger battery mounted beneath it. Half the volume of the A3019A is taken up with its 48 mA-hr battery. Three quarters of the volume of the A3019D is taken up with its 225 mA-hr battery. The larger battery allows a more efficient encapsulation process, so we take up less volume with epoxy. The result is an unattractive but long-lasting transmitter.

Figure: Subcutaneous Transmitter A3019D with Screw Electrodes. This transmitter is designed for use in adult rats. Volume is 2.4 ml, diameter 23 mm. Operating life is a minimum of 16 wks at 512 SPS, with typical operating life of 18 wks. The white lead is the X+ terminal. The antenna is stainless steel in a tough silicone tube for endurance in an active rat. The helical steel leads are insulated with five coats of silicone.

We determine the version of a transmitter during assembly, programming, and encapsulation. We might change capacitors on the board to set the filter frequencies. We might change resistors to set the gain. We program the logic chip to set the sample rate for each channel. We can equip the circuit with a small, medium, or large battery. The larger the battery the longer the operating life of the transmitter, but the greater its volume. The following table gives the versions of the A3019A that we have defined to date. Click on the version names to see pictures of the individual versions, should they exist.

Versions C and D look the same from above, but the C version uses a slightly thinner battery, for a maximum height of 6.4 mm instead of 7.0 mm. For comparison, the A3013A, which we implanted in dozens of rats with success, was 17 mm × 19 mm × 10 mm. Its volume without its antenna or leads was 2.4 ml.

By default, the transmitters come with three coats of silicone. But the "-T" suffix specifies five coats, and these transmitters are typically 1 mm thicker than those with only two coats. Each silicone coat applies another 250 μm of silicon, so considering both sides, we add a total of 1 mm with two coats.

Design

S3019_1.gif: Subcutaneous Transmitter Circuit Diagram.

Code: Logic Chip Firmware Library.

A301901B.zip: PCB Gerber Files.

A301901B_Panel.zip: PCB 2×5 Panel Gerber Files.

A3019A.xls: BOM, PIK, KIT, and Cost for A3019A Production.

R3019.xls: Calibration currents and battery dates for A3019 transmitters.

B22260_SMT.pdf: Component placement drawing from Advanced Assembly.

B22260_KitLab.pdf: Kit labels from Advanced Assembly.

B22260_AB.pdf: Parts list from Advanced Assembly.

Analog Inputs

The two input leads are X+ and X−. The antenna is OUT. You can see all three signals declared with these names in the circuit diagram. These are the only three connections we make to the circuit. The antenna is insulated from the outside world by a silicone tube. It transmits the value of X with radio waves. The X+ lead is white. The X− is transparent, showing the coils of the steel helix inside.

The A3019 amplifies and filters X to produce a signal XA which passes into a sixteen-bit ADC. The result is a sixteen-bit number. The number is 0 when XA = 0 V and 65535 when XA = VBAT, the battery voltage. With a fresh battery, we have VBAT = 2.7 V. The A3019A has gain 100 in their amplifier, so we see that each ADC count represents roughly 400 nV at the X input. In the last few days of a transmitter's life, the VBAT drops to 2.2 V, at which point the transmitter fails. During this time, the voltage represented by each ADC count drops in proportion to the battery voltage. When VBAT is 2.2 V, one count represents roughly 350 nV.

Aside: By comparison, one ADC count in the earlier A3013A transmitter represented 140 nV at the X input. The gain of the A3091A is three times greater than that of the A3013A.

In theory, the dynamic range of the A3019A is 2.7 V ÷ 100 = 27 mV. But in practice, the amplifier is not able to drive XA all the way up to VBAT or all the way down to 0 V. So we quote the dynamic range as a more conservative 20 mV, or if we want to be more specific, −14 mV to +6 mV.

We discuss the noise present on the X input in our presentation of the A3020 prototype circuit here. With gain of 100, we expect less than 30 counts of noise in a 160-Hz bandwidth, which is 12 μV.

Like the A3013, the A3019 provides a DC-blocking filter on the X input as soon as it enters the circuit. The average value of XA is VCOM, which is 1.8 V. Thus the average value of the digitized signal corresponds to 1.8 V, and from this we can deduce the current value of VBAT. We used the average value of X to monitor battery voltage in the A3013. The same calculations apply to the A3019, regardless of the gain of its amplifiers.

The impedance between X+ and X− is 10 MΩ in series with 100 nF. The antenna makes no low-frequency connection with the outside world. It transmits radio-frequency waves out through its silicone sheath. The X input is isolated in that its associated amplifiers are not connected to the outside world at all, except through the capacitance between the transmitter battery and the animal body, which is a few tens of picofarads.

The A3019's analog filters and amplifier are similar to those of the A3013, which we describe in detail here. The principle difference between the two circuits lies in the elimination of one of the three op-amps used in the A3013 to arrive at a more compact circuit for the A3019. The input impedance of the two circuits is identical: 10 MΩ in series with 100 nF. But in the case of the A3019, the high-pass filter is dominated by the combination of R10 and C9. With 50 kΩ and 4.7 μF, the half-power frequency of this single-pole high-pass filter is 0.68 Hz.

Resistor R10 also sets the gain of the amplifiers, as shown by a table in the schematic. With R10 = 50 kΩ, the gain is 100. With R10 = 100 kΩ, the gain is 50 and the high-pass filter frequency drops to 0.34 Hz.

The A3019's three-pole low-pass filter circuit is identical to that of the A3013A. The following graph shows the measured response of this low-pass filter with frequency for various cut-off frequencies. We set the cut-off frequency by our choice of the filter capacitor value CF, as shown in the table in the schematic.

Figure: Frequency Response of Low-Pass Filter. We apply a 1-mV amplitude sine wave of increasing frequency to the input and record the rms ADC output in ADC counts.

The plots are normalized with respect to the response at 0.5 Hz in the presence of a high-pass filter with half-power frequency 0.17 Hz. When we make the same measurements for the A3019A circuit, we see the high-pass filter coming into play below 1 Hz. The following graph shows the measured gain from X to XA through the A3019A amplifiers and filters combined.

Figure: Measured Frequency Response of A3019A amplifier and filters.

The above graph is taken from A3020 prototype board manual, where you will find more details of the circuit performance.

Input Noise

[06-MAY-11] With electrodes shorted together, or when immersed in pure water, most A3019As generate 12 μV of noise in their 160-Hz pass-band. The following table gives the amplitude of the 2-20 Hz and 60-160 Hz bands for fourteen A3019D transmitters in cool and warm water. We measured noise in salty water and pure water and found that the addition of salt had no effect. The table includes the baseline EEG amplitude we obtain from these same bands using eight transmitters implanted in control animals, as we describe here.

In cool water, noise power is three or four times lower than baseline 2-20 Hz EEG power. In the 60-160 Hz band, 5.6 and 5.10 produce significantly more power than EEG. In warm water, however, the noise power drops to the levels we are accustomed to observing for unencapsulated circuits with their inputs connected together.

We find that noise power in the 2-20 Hz and 60-160 Hz bands increases in proportion to Hall Effect Switch noise. In cool water, 5.10 has a 5-μV peak at 18.5 Hz, as you can see here. In warm water, this peak disappears, as you can see here.

[08-MAY-11] Further experiments show the Hall Effect Switch noise arising for temperatures below 25°C, with the 5.10 peak reaching a maximum amplitude of 9 μV at 15°C, and decreasing for lower temperatures. The A1171 Hall Effect Switch (U1) uses an internal switching frequency of around 200 kHz to obtain a measurement of local magnetic field strength roughly 18 times a second. The noise we observe is present even when the input to our amplifier is shorted. At the same time, its amplitude starts to fall off above 160 Hz, although not as sharply as if the noise were inserted at the input to the low-pass filter. From this we conclude that the noise is entering first stage of the amplifier, which is equipped with resistors of order 1 MΩ. If we assume a capacitance, C, between the switching voltage and a 1-MΩ pick-up impedance in the amplifier, and we assume internal switching the full 2.7-V battery voltage at 200 kHz, we expect the 200-kHz amplitude on our input to be around 1 mV/fF (one millivolts per femto-farad, which is one thousandth of a picofarad).

Any non-linearity in the first stage of amplification at 200 kHz will demodulate the switching frequency, giving us a signal that follows the amplitude envelope of the switching frequency. For an A1171 that turns on at 18 Hz, we will see a peak at 18 Hz, harmonics at multiples of 18 Hz, and plenty of other noise at higher frequencies, which indicate the manipulation of the switching frequency by the A1171 during its measurement intervals.

Suppose C consists of two 1 mm square plates separated by 5 mm. One plate is the U1, the other is U8 with its surrounding traces and resistors. The value of C in air will be around 2 fF, and in epoxy 10 fF. The induced 200-kH noise will reach amplitude 2 mV or 10 mV. The demodulation efficiency need be only 0.1% to give rise to the noise we observe.

Significant switching noise occurs only when the circuit is encapsulated in epoxy. The epoxy can be liquid or hardened, it makes no difference. The permittivity of epoxy is over four times that of air. We can imagine the value of C increasing by a factor of four when we encapsulate, increasing the 200-kHz coupling to the point where the demodulated signal is significant.

The demodulation of a 10-mV, 200 kHz signal will take place within the non-linear elements of the op-amp. The efficiency of such demodulation could be a strong function of temperature, which could explain why we see the switching noise is such a complex function of temperature. A corollary of this hypothesis is that the size of the noise on X is a function of the op-amp not the hall switch.

Body Capacitance

We use X+ or X− as as a differential input. If you have a ground potential and an action potential, you can use either input as your ground. You never need a third wire for a low-impedance ground. Indeed, there is nowhere to connect such a third wire, and any effort to add such a connection would undermine the sensitivity of the X input.

The X− input is connected to the VCOM power supply. When biometric voltage source moves the electric potential of X− up and down, the potential of the transmitter circuit and its battery moves up and down also. If VCOM goes up by 1 mV, so do the VBAT, 0V, and 1V8 voltages in the circuit. The charge the voltage source must push into the transmitter to change its potential by voltage V is equal to CV, where C is the electrical capacitance between the transmitter circuit and the animal body, which we call the transmitter's body capacitance.

The capacitance between two parallel conductors of area A separated by a distance d is Aε/d, where ε is the permittivity of the insulating medium. Consider the A3019D with its 25-mm diameter battery. This battery is coated with roughly 500-μm of epoxy and silicone, which have permittivity 40 pF/m. The A3019D body capacitance will be roughly 40 pF. The A3019A body capacitance will be around 10 pF. The capacitance of a neural membrane is of order 1 μF/cm², with surface area of brain neurons around 0.003 cm², so that the capacitance between a neuron and its surrounding fluid, which we shall call the neuron's body capacitance, is of order 3 nF.

A neuron's body capacitance is one hundred times the body capacitance of our transmitter. Suppose we connect X− directly into a neuron with a hypothetical, tiny, low-impedance probe. When the neuron generates a potential, it will have to push a hundred times more charge into its own body capacitance than it will push through X− into the transmitter's body capacitance. If we use a larger probe to measure the average voltage of a hundred neurons, their membrane capacitances will be acting in parallel, and once again, the transmitter's body capacitance will be insignificant. When we connect X− to a biopotential source, and leave X+ isolated, the source will not notice the presence of X− at all.

Meanwhile, our X+ input does not connect to the transmitter power supplies, and so is not subject to the transmitter's body capacitance. When we connect X+ and X− across a biopotential source, the source will see 10 MΩ in series with 100 nF. The 100 nF presents an infinite impedance at 0 Hz. Above 1 Hz, its impedance is small compared to 10 MΩ.

We conclude that in any biometric application, it matters not at all whether we connect the X+ or X− lead to our probe, other than that the polarity of the signal we observe on the screen will reverse when we reverse the leads.

Almost all the surface area of a neuron is provided by dendrites and the resistivity of dendrite membrane is roughly 50 kΩcm². With total area 0.003 cm², the resistance from the neuron body to the surrounding fluid through its dendrites will be 20 MΩ. In parallel with this resistance is the neuron body capacitance of roughly 3 nF. The two in parallel are the neuron's body impedance.

Suppose we apply our X input directly across the body impedance of the neuron using two hypothetical, tiny, low-impedance probes. A 0-Hz signal generated by the neuron will not notice the transmitter's input impedance, because the input impedance is infinite at 0 Hz. At 100-Hz, however, the input impedance becomes 10 MΩ, so that a 100-Hz signal generated by a neuron will be loaded by its 3-nF body capacitance as well as 10 MΩ. At 100 Hz, 3 nF has an impedance of 500 kΩ, which is far smaller than our 10 MΩ. At 0 Hz and above 100 Hz, the input impedance of our transmitter is insignificant. At 1 Hz, however, the neuron body capacitance has impedance 50 MΩ. In parallel with this impedance is the 20-MΩ membrane resistance and the 10 MΩ transmitter input resistance. So at 1 Hz we see that the application of the X input to the neuron will more than double the current flowing out of the neuron for a given signal voltage. We can expect this increased current flow to cause a decrease in the signal amplitude.

Suppose we measure the potential within such a neuron with a thin probe tip whose front-end impedance is 100 MΩ. Our 10 MΩ input impedance will reduce the potential at the back end of the probe. The 10 MΩ resistor in the transmitter will draw current out of the neuron and this current will cause a voltage drop between the probe tip and the transmitter input. The potential at the front end of the probe will be unaffected because the resistance within the neuron is small compared to the front-end resistance of the probe tip. But the 10 MΩ input resistance of the A3019 will attenuate the signal at the back-end of a 100-MΩ probe.

Suppose measure the average potential of a hundred neurons by placing two large low-resistance electrodes, such as a stainless steel screws, at two places in the fluid between the neurons. The source resistance of this voltage will be many 20-MΩ resistors in parallel, as well as many 3-nF capacitors in parallel. Our 10 MΩ input impedance will be much larger than the source impedance, and will have no effect upon the size of the voltage we measure. It is such biometric potentials as these that the A3019 is designed to measure.

Battery Life

The A3019 contains an A1171 hall sensor switch, U1 in the schematic, and an SN74AUP1G80 D-type latch, U2 in the schematic, to turn on and off power to the transmitter when we apply a magnetic field. The switch and latch together consume 5 μA from the battery at all times. Thus the inactive quiescent current of all versions of the A3019 is 5 μA.

The shelf life of a transmitter is the length of time it can sit inactive on the shelf before its battery is exhausted by the inactive quiescent current. With a 48 mA-hr battery, such as the BR1225 used in the A3019A, the shelf life is 9600 hours, or 57 weeks. With a 255 mA-hr battery, such as the BR2330 used in the A3019D, the shelf life is six years. In this respect, the A3019 is very much superior to the A3013, whose inactive quiescent current was around 22 μA. Indeed, the 1.0-ml A3019A transmitter would not be practical with a 22-μA inactive current.

The active current consumption of the A3019 consists of the quiescent current of the logic chip, which is independent of sample rate, and current consumption of the sample and transmit process, which increases linearly with sample rate. We studied the current consumption of the A3019 with the help of the A3020 test board, as we describe here.

After assembling twenty A3019As, we arrive at the following empirical formula relating the active current consumption to the sample rate, M. The formula agrees with our observation that the active current consumption is, on average, 80 μA, with a standard deviation of 4 μm.

I_active = 24 μA + (M × 0.11 μA)

The operating life of a transmitter is the length of time it can run before its battery is exhausted, assuming it started with a fresh battery. With a 48 mA-hr battery, the operating life of a 512 SPS transmitter is 3.6 weeks. With a 255 mA-hr battery, the operating life is 19 weeks.

Encapsulation

Warning: The A3019 circuit board is only 0.8 mm thick. It is so flexible that we must take great care not to flex it when breaking it out of its panels, soldering wires to it, and cleaning it. If we bend it, capacitors and resistors on the board are likely to break. The danger of flexing it greatest when the programming extension is still attached.

For all versions of the A3019, we coat the antenna with silicone directly, instead of using a silicone tube. The result is a tougher antenna, and protection for the radio-frequency oscillator during construction.

Encapsulation of the A3019A is almost identical to Process B, which we developed for the A3013A. The A3013 process reached its final state on 18-AUG-10 when we encapsulated transmitter 5.82, the last of the A3013s. The A3019A has 60-mm leads, designed for use in mice, with only four coats of silicone instead of five (see here).

The encapsulation of the A3019D does not involve total immersion of the device in epoxy. The fact that the A3019D's battery is twice the area of the circuit board means we can use a process that does not require immersion, and if properly executed, we avoid all clipping and filing of the epoxy. Here is the A3019D encapsulation process in detail.

1. Prepare Battery: Apply two coats of black nail polish to the metallic top surface of the battery and the plastic outer edge.
2. Prepare Wires Take two 150-mm helical stainless steel wires. Hold one end with a clothes pin. Put on a pair of plastic gloves. With your fingers, extract some silicone dispersion from your reservoir and place it upon a piece of paper. Pick up silicone with your fingers and drag the wire through once to apply silicone and a second time at a speed of 10 mm/s to smooth out the coat and remove excess. Hang by the clothes pin over hot water for half an hour. Apply a total of five coats coats of silicone in this way. In the case of white leads, make the fifth coat a white coat. Allow to cure for twenty-four hours over warm water after the final coat. Coat 150-mm lengths of antenna in the same way, with five coats, and later cut 52-mm sections off the coated wire for the transmitters.
3. Test Switch: Connect circuit to current monitor with battery. Short R1 with tweezers to force start-up of magnetic switch. Check that magnet switches current off to 5 μA.
4. Test Logic: Apply shorting jumper between P3-1 and P3-2. Connect logic programmer to P1. Turn on power. Program chip with firmware that enables test pins and assigns a new channel number. Disconnect programmer. Current should be between 85±5 μA for all versions that operate at 512 SPS. Connect probe to P3-4 and check 32.768 kHz square wave. Turn the circuit on and off with magnet a few times to confirm operation of switch and start-up of oscillator.
5. Calibrate Ring Oscillator: Connect probe to P1-3, turn on circuit. Observe transmission burst. Measure period of start-up square wave. Should be 200 ns. Adjust tcd_divisor and tcd_half_divisor to set period and mark-space ratio. Compile and re-program. Measure again.
6. Calibrate VCO: Place the antenna of a Spectrometer (such as the A3008) up against the antenna hole on the circuit. Obtain a spectrum. Adjust frequency_lo in the firmware to place the center of the spectrum between 914 MHz and 918 MHz. Compile and re-program. Measure again.
7. Final Program: In firmware code, turn off the test pins. Compile and re-program. Measure current consumption. Should be 80±5 μA. Record serial number, channel number, frequency_lo, tcd_divisor and tcd_half_divisor in spreadsheet.
8. Check Reception: Place the antenna of a Data Receiver (such as the A3018) up against the antenna hole on the circuit. Check reception with the Recorder Instrument. The number of messages received per second should be over 90% of the nominal rate for the transmitter (512 MSP for A3019A). The analog input should show a fifty to a hundred counts rms mains hum.
9. Attach Wires: Choose two silicone-coated leads. Cut to the required length length. With a scalpel, cut around the circumference of the silicone 2 mm from each tip of both leads. Turn the detached 2-mm silicone segment so as to rotate it off the thread of the lead. Solder leads into the X+ and X− holes in the circuit board with the help of water-soluble acid flux. Punch two holes in a piece of cardboard and insert two screws for the electrodes. Tin one half of the top of the screw with the help of acid flux. Clip the free ends of the leads to 1 mm. Tin them with acid flux. Solder lead tips to the screws. Cut a 52-mm length of antenna wire. Remove silicone insulation from final 2 mm of one end with a scalpel. Tin this exposed tip with acid flux, then solder into circuit's antenna hole with the help of more acid flux. Let the circuit soak in hot water for ten minutes. Clean with hot water and a brush, taking utmost care not to bend the circuit board. Blow dry with compressed air. Bake at 60°C for one hour to evaporate all remaining moisture from beneath the components.
10. Check Input Response: Plug a battery into the transmitter. Activate transmitter and observe signal with Recorder Instrument. Wave the transmitter around. We expect to see mains hum with amplitude varying between 10% and 100% of the transmitter's dynamic range. Press the X wire tips together. The signal should settle to a constant value with standard deviation 15 to 30 counts. Hold the tip of one lead in each hand and allow the transmitter to come to rest. You should see pulses in the trace that correspond to your heart beat. We call this the heartbeat test. Assuming you get mains hum and heart beat, the circuit is ready for encapsulation.
11. Load Battery: Hold the electrode leads and the programming extension in one hand. Clip off the transmitter extension. Do not hold the transmitter circuit while you clip, because the shock may damage some of the components. Solder the battery to the circuit. Use no-clean solder and wipe off excess flux with ethanol. Make sure there are no sharp points sticking up from the joints or wire ends. Examine the transmitter with a loop to make sure there is no contact between the battery leads and the circuit or the input leads. Test the transmitter with the Recorder Instrument. Cover the exposed VBAT solder joint with fast-drying nail polish.
12. Encapsulate: Hold the transmitter in one hand and apply EP965 epoxy to the edges of the circuit board, the top metallic surface of the battery, and the top surface of the circuit board. Place battery-side down in a petri dish with a teflon floor cover. The room temperature should be 20±4°C for the epoxy flow and curing to cooperate properly for the first fifteen minutes after we apply it to the circuit. Put the petri dish in vacuum for two minutes. Fill the chamber with air. Evacuate a second time and allow to sit for another two minutes. Re-fill the chamber. Remove and test the transmitter by applying a magnet to the underside of the dish and observing signal with Recorder Instrument. Allow to cure.
13. Touch-Up: Touch up the epoxy with black nail polish. Apply multiple coats if necessary, to make sure the battery leads and all components are covered.
14. Bind Wires: Bend the antenna back on itself and fasten to the X leads with a 5-mm length of heat shrink that begins 5 mm from the transmitter body and ends 10 mm from the transmitter body. The tip of the antenna should be flush with the edge of the heat shrink. The entire assembly should be such that when we hang the transmitter hangs by the X leads, the VBAT terminal is the lower-most point on the transmitter. By this means, the VBAT terminal will be well-covered by the drop of silicone that forms during each silicone coat. Test the transmitter with the Recorder Instrument.
15. First Silicone Coat: Dip the transmitter in silicone dispersion. Allow the antenna to sink into the silicone. Hang the transmitter by its wires over a tray of hot water. Leave to cure for half an hour.
16. Apply Label: Apply an identification label to the top surface of the battery.
17. Second Silicone Coat: Dip the transmitter again and hang over hot water for half an hour.
18. Third Silicone Coat: Dip the transmitter again and hang over hot water for half an hour.
19. Fourth Silicone Coat: Dip the transmitter again and hang over hot water for half an hour.
20. Fifth Silicone Coat: Dip the transmitter again and hang over hot water for 24 hours.
21. Trim Silicone: Inspect the silicone. Cut off the cured drop around the VBAT terminal. Make sure the base of the antenna and the X leads is well-coated and flexible.
22. Check Input Response: Collect a batch of transmitters. Turn them on and hold the tips of their X+ leads together. Lower the bodies and X− leads in a beaker of warm water (>25°C). Look at the transmitter signals with the Recorder Instrument. They should be almost identical, as they are here. This is the harmony test. Drop the transmitters into the beaker. All transmitter signals should have standard deviation less than 30 counts. Amplitude of all peaks in the 15-20 Hz band should be less than 1 μV. This is the noise test. Allow the water to cool to room temperature (<20°C) and look at the 15-20 Hz band again. All peaks should be less than 10 μV.
23. Leak Test: Put the batch of transmitters in a vacuum four five minutes. Remove, turn on, and drop in water. Leave running in water for four days. At the end, repeat the harmony and noise tests.

When we allow epoxy to cure, we mean that we leave it for 24 hours at room temperature, or we leave it for one hour at room temperature followed by one hour at 60°C. The initial hour at room temperature is important, because warm uncured epoxy is less viscous and runs off the objects being coated. Be sure the epoxy is not too cold when first applied, because if it is too cold it will not cure fast enough, and it will continue to creep off the battery and onto the base of the petri dish.

Modifications

The following modifications are required by the A301901A circuit board.

1. Cut track from U6-A3 to U4-1.
2. Connect U4-1 to U4-4 with a wire.
3. Connect scope probe ground pad to P2-1.

The A301901B requires no modifications.

Development

[07-APR-10] We receive our A301901A circuit boards and start building four boards by hand. We have difficulty with U1, the A1171 in a DFN-6 package. We can solder the package by applying no-clean BGA flux to the board, tinning the footprint pads, placing the DFN-6 on the footprint, and heating the chip from above with a soldering iron and a blob of solder as a conductor. With the soldering iron at 350°C we find that the A1171 magnetic switch provides erratic performance. We become confused between the quality of the soldering and the operation of the chip. All chips that functioned reliably afterwards, through heating and cooling from further assembly and cleaning, we applied with the soldering iron at 300°C. One mode of failure was excessive current consumption, bringing down the junction of R1 and C1 to the point where the chip was improperly supplied. We suspect, but we are not certain, that the A1171 is particularly sensitive to heat, in the same way that we previously observed with the LC4064ZC in a BGA-56 package.

Having figured out how to get the DFN-6 on the board, we now worked on the BGA-5 of U2. We used roughly five chips for every one we mounted correctly. It is easy to smear the balls under the chip. But these chips do not appear to suffer from damage by excessive heat, so once they are on the board, they are reliable.

On 29-MAR-10, with 200 of our A301901A printed circuit boards in fabrications, we received notification from DigiKey that the ASHK oscillator by Abracon had been discontinued by the manufacturer. We immediately attempted to buy 100 of the parts to tide us over, but we were too late. The only drop-in replacement we could find is the ECS-327KO by ECS, which will operate between 1.3 V and 5.5 V, although it is specified at 3.3 V. In our circuit it receives 1.8-V power. We bought 100 of these. They worked fine in the A3020A circuit, where we tied the tristate input U4-1 to the power supply U4-4. In the A301901A layout, we connect U4-1 to U6-A3. Pin U6-A3 is in the 3-V I/O bank of U6. It provides only 3-V or 0-V for its output. The ECS-327KO data sheet assures us that we can apply 3 V to the tristate input even when the power supply is 1.8 V. We confirmed this with ECS. So we figured all should be well.

We applied 2.7 V to U4-1 with U6-A3. This 2.7 V is the actual voltage on the 3VD power supply in our battery-powered tests. Current consumption increased by roughly 15 μA. With U6 programmed to HOLD its input levels, which is necessary for low current consumption, U4 does not oscillate. When we switch to PULL-UP inputs, oscillations start again. We conclude that U4 cannot drive a HOLD input when its tristate input is being driven with a voltage greater than its supply.

We cut the track from U4-1 to U6-A3 on three A3019A circuits. After a thorough cleaning, and assuming we don't short U4-1 to 0V with a probe, U4 provides 32.768 KHz. We conclude that we must cut this track on every A301901A circuit board.

Current consumption when OFF is 4.0 μA. We find that we can reduce current consumption when ON by 1.3 μA if we disable the test pin outputs.

By end of day, we have three working A3019As without amplifiers. All are programmed to digitize VCOM, with ID numbers 1, 2, and 3. No3 has a curious problem whereby we must short R2 or else current consumption jumps to 500 μA.

During the course of the day we replaced all logic chips at least once. We applied new chips by heating the board from underneath for two minutes, then blowing hot air from above for thirty seconds. Current consumption when active is between 70 μA and 80 μA. We recall problems with varying current consumption in the A3009A, where we were damaging the logic chip with our heat gun. We left two of the transmitters to run overnight.

[08-APR-10] Current consumption of our three transmitters is No1: 73 μA, No2: 80 μA, and No3: 69 μA. Noise on the analog signal is No1: 13 counts, No2: 15 counts, No3: 57 counts. We encounter no problems with U4 starting up when we turn on power. We add analog amplifiers to all three circuits. Current consumption is now No1: 73 μA, No2: 83 μA, and No3: 72 μA. Reception from all three is robust. We leave them transmitting.

[22-APR-10] We encapsulate two dummy circuits, an A3019A with the 48-mA-hr battery and an A3019B with the 120-mA-hr battery. Both require care to cover the protruding battery leads, but encapsulation proceeds well and the result is two transmitters with three coats of silicone each and volume 1.0 ml and 1.8 ml respectively.

[07-MAY-10] We receive nine A3019A from Advanced Assembly in Colorado. We expected a full panel of ten, but one board was "burned in re-work". The printed circuit around the missing board in the panel was scorched. The remaining boards had problems, as we describe in this e-mail. The A1171 chips on eight of the boards did not work. We re-mounted them and six of them worked erratically. So we think they were damaged by heat, but we cannot be certain because we have no fresh chips with which to replace them. The other two worked reliably. We ended up with three working circuits out of nine. We discovered that we could re-mount the BGA-5, U2, after taking its balls off, which we found much easier than mounting it with the solder balls in place.

During the course of our tests, we found that U4 can fail to start up properly when its enable pin, U4-1, is floating and we have the programming jumper across R2 removed. With the programming jumper, the chip starts up reliably, provided that you have not dragged the enable pin down to 0V with a scope probe in the previous two minutes, and provided that the board has been well-cleaned and dried. We now understand why our previous hand-made board No3 consumed excessive current without shorting R2. The oscillator consumes excessive current when it starts up, pulling down the 1V8 supply by the voltage drop through R2. With U4-1 floating, the oscillator does not start properly and enters an unstable state that continues to consume current. We connect U4-1 to U4-4 so as to pull-up the enable line, and now we find that U4 always starts up, even with the initial voltage drop across R2, and furthermore we can look at U4-4 with a scope probe and have no effect upon the voltage on the enable pin. We have a choice of solutions to the start-up problem: replace R2 with 0 &Ohms; or apply a wire link to U4. We prefer to add the wire link. If we remove R2 we are left with the concern that noise from the 1V8 supply will contaminate VCOM, which we don't notice now with a gain of 100, but we may notice later with a gain of a lower gain. And with U4-1 floating, we have an unpleasant problem with the pin staying low even after we cycle the power.

[13-MAY-10] We receive new A1171 chips and replace those that we thought were damaged by heat. We see precisely the same erratic behavior. The voltage at U1-6 is around 0.8 V and U1 is unresponsive to a magnet. When we short R1 with tweezers, U1 starts up and works fine. We conclude that most A1171 chips will fail to start up most of the time when they are powered through a 50-kΩ resistor. We can either drop the value of R1 or short it when we connect the battery. If we drop R1, we will introduce more asynchronous noise onto VCOM from the A1171's internal switching clock. If we rely upon shorting R1, then we must trust that the battery power will be continuous during the life of the encapsulated circuit.

After three days of drying, two circuits with power consumption >130 μA are now consuming less than 90 μA. Another circuit that showed excessive consumption has some kind of printed circuit board error. We removed all parts from the board except the magnetic switch and still we see 300 μA consumption.

We retrieved transmitter No3 and found that its U4 was having the same start-up problem we observed elsewhere. But connecting U4-1 to U4-4 did not solve the excessive power consumption problem with R2 = 1kΩ. We see a step down of 0.5 V on 1.8V during transmission. The rising and falling edges of this step have time constant 1 μs, which suggests R1 with C5 but in the absence of C4. We replaced C4. For some reason C4 is not connecting to the 1.8V supply. Consumption drops from 350 μA to 75 μA when we short R2 and we see no increase in input noise.

We now have 7 working boards out of the 9 sent to us by AAPCB. We decide to proceed with production with the circuit as-is. We will start the circuit by shorting R1 when we install the battery. We will apply a wire link to U4, which takes less than 5 minutes per board, including cleaning. The result will be a robust circuit with the minimum noise injected by the logic and switch circuits into the analog channel.

[01-AUG-10] We received another 61 boards from AAPCB. We tested a few at random and they appear to work well. We damaged three of them by connecting power the wrong way, but we used these in dummy encapsulation experiments, which was a valuable exercise. We shipped one A3019A to Oxford and gave another one to CHB. We expect the first EEG recordings with the A3019A soon.

[18-AUG-10] We prepare the first example of the A3019D using a faulty circuit, to get an idea of its volume and appearance. This transmitter has the same volume as the earlier A3013A, but is thinner and wider. We hope that is will be more suitable for implantation, and at the same time will provide 19 weeks of operating life rather than 9 weeks.

[23-SEP-10] At CHB they have recorded seven hours of data from an implanted A3019A with channel No1. Reception robustness was 95% but the electrodes appear to be poorly attached (see here). Meanwhile, Louise Upton reports that the A3019A will fit in a mouse. We delivered an A3019D to ION and await results.

[06-DEC-10] We encapsulated 9 A3019Ds using a procedure closely modeled after Process B for the A3013A. The transmitters end up looking like this and have volume 2.6 ml. We them after each stage of encapsulation. Two failed after clipping off the excess epoxy. Because we had been careless with excess epoxy, the clipping was particularly forceful, causing pain the hand that held the transmitter. We suspect that the shock of the clipping damaged the assembled circuit or broke the solder bridge connection between the battery positive terminal and the circuit board. We try out another process, where we cover only the circuit board and top side of the battery with epoxy, cure, and cover the bottom of the battery with a film. This requires two immersions in vacuum, but no clipping or filing. We use a short wire to connect VBAT to the circuit.

[08-DEC-10] After four coats of silicone, the transmitter is 23 mm in diameter, with volume 2.2 ml, compared to 25 mm in diameter and 2.6 ml for the previous process. The encapsulation appears rugged, and we were pleased to avoid the clipping and filing steps. We will now attempt to encapsulate seven functional transmitters this way.

[13-DEC-10] Seven transmitters encapsulated with new method, no failures. Had to deal with the VBAT wire protruding through the epoxy, so added a drop of silicone on the wire for each transmitter. Result is satisfactory. Also tried two types of nail polish on wires and an aluminum plate with coat of silicone after. Found that the quick-dry polish held well to the wire, but normal polish seeped off. Quick-dry polish adhered well to aluminum, normal polish peeled off easily. Silicone adhered well to both polishes.

[14-DEC-10] Washed all seven new transmitters. Placed in vacuum for ten minutes. Observed no swelling beneath the silicone. Placed in warm salt water and turned on. With water recently stirred, standard deviation of signals from all seven is between 30 and 100 counts. Transmitter No5 is hard to turn on and does not pick up mains hum properly. It shows many bad messages in reception and curious pulses of a millivolt or so. Leave running in the saltwater.

[15-DEC-10] Transmitter No5 is having fewer jumps, but does not pick up mains hum well, and is noisy. We have trouble turning it on and off. We remove it from the batch and bend up the base of the leads to look at the vulnerable base of the leads. We see a tear. We wash with water, wash with alcohol, dry, and apply silicone to the tear. This place is vulnerable because it is more flexible than the heat shrink or the body. In the transmitters where we forced the heat shrink all the way down to the transmitter body, there is no such weakness.

[16-DEC-10] We put the 7 transmitters we encapsulated with the original clipping process in vacuum for ten minutes, then in water. They are all working well. At the end of the day we take them out and dry them. We hang up all 13 transmitters and apply a coat of silicone to the vulnerable section of the lead base.

[17-DEC-10] No5 is working well now. It picks up mains hum, is no longer noisy, and switches on and off well. We leave it in water. All other transmitters working well. We apply another two coats of silicone today and take a photograph.

[18-DEC-10] No5 turns on and off easily, but its analog input is noisy and no longer picks up mains hum. We inspect the silicone with a loop and see no sign of a tear. The two coats of silicone we applied to the previous tear are still intact. We do not understand the intermittent failure of No5's analog inputs.

[21-DEC-10] We complete the first A3019C. We use red nail polish to cover the VBAT connection before we apply silicone. We cover the back of the battery before we solder it to the transmitter. We used so much glue that there is a 1-mm bulge across the back. The volume of the A3019C volume is 2.2 ml instead of our hoped-for 1.9 ml. This transmitter is No6. Transmitter No5 has been in water, and remains the same.

[07-JAN-11] Six new A3019Ds have been in water for twenty-four hours and are working well. They look like this, with the heat shrink pressed up against the transmitter body and the antenna on the underside. The underside is weak, and we have reinforced with silicone.

We perform the harmony test and see this. The agreement between the transmitters for a variety of step changes, 60 Hz hum, and in the amplitude of the harmonics of 60 Hz, is conclusive evidence of proper function in the analog amplifiers and filters. We perform the heartbeat test and see heartbeat in each case.

We apply the third coat of silicone to seven A3019Ds. These are the first we made with the heat shrink 6 mm from the transmitter body. The result is no weak point in the joint between the leads and the body. We still take the precaution of touching up the base of the leads and the antenna with silicone, and the battery terminal, but we do so only once between the second and third coat. The result is satisfactory and compact.

[22-APR-11] We measure current consumption in transmitters with antennas attached. For six such transmitters, we find the current consumption is 3 μA higher than we measured without the antenna. We conclude that the antenna adds 0.6 mA to the current consumption of the RF oscillator, for 1.6 mW of increased power dissipation during transmission. Of this, we have in the past measured a transmitted power of around 0.5 mW. Thus when we calculate battery life, we must add 3 μA to the active current, and this decreases the average battery life for A3019D transmitters from 19 wks to 18 wks.

[04-MAY-11] Transmitters 5.1-5.13 and 4.14 are encapsulated. These are destined for Rob at UCL. They pass the harmony and heartbeat tests. We place them in water, with their leads mixed up and separated. We look for noise from the Hall Effect Switch, which will show up as a peak between 15 and 20 Hz.

Figure: Hall Effect Switch Noise on Newly-Encapsulated Transmitters. The fourteen transmitters are totally immersed in water, and have been for no more than an hour. Signals recorded in M1304532983.

Transmitters 5.8, 5.9, and 5.13 show no sign of a peak at any frequency. The others show peaks between 15 Ha and 20 Hz. The peak amplitude is 3 μV to 8 μV. Total noise amplitude in 0-160 Hz varies from 10 μV rms for transmitters peaks below 4 μV to 28 μV rms for 5.10 with its peak amplitude of 8 μV. The following plot compares the full-spectrum noise of three transmitters.

Figure: Hall Effect Switch Noise from Three Selected Transmitters, 5.1, 5.9, and 5.10. Signals recorded in M1304532983. Noise in 0-160 Hz is 11 μV, 10 μV, and 27 μV rms respectively. The noisiest transmitters are 5.3, 5.5, and 5.10.

The 5.10 transmitter has an 8-μV peak in the switch noise. We see noise throughout the pass-band of the A3019D's 160-Hz low-pass filter. Above 160 Hz, the noise attenuates, but not as sharply as it would if it were subject to the full attenuation of the three-pole 160-Hz low-pass filter.

[05-MAY-11] We go back and look at M1294435389, recorded on 07-JAN-11, which shows the signal from 3.2, 3.6, 3.7, 3.8, 3.10, and 3.12, which we sent to Rob at UCL soon after. At time 60 s we see the input noise when the transmitters are immersed in water. We see switch noise on transmitters 3.7 (16.25 Hz, 2.5 μV), 3.8 (17.0 Hz, 2.5 μV), and 3.12 (16.125 Hz, 1.7 μV). Transmitter 3.8 subsequently failed with 1-Hz oscillations.

We check the 14 transmitters immersed in water and obtain this plot of switch noise, which is much the same as yesterday's. We try transmitters 5.14 and 6.1, which are A3019As freshly encapsulated. Number 5.14 shows a 3-μV peak at 15.5 Hz, while 6.1 shows a 1.5-μV peak at 18.25 Hz, as you can see here. They pass the heartbeat and harmony tests. With inputs shorted together or in water, noise is around 12 μV from both.

We have seven functioning unencapsulated transmitters with programming extension, no antenna, and no leads. We power each one in turn with a battery on a connector, check the Hall Effect Switch is working, and look for switch noise. We see none. We are certain the switch noise is less than 1 μV in all cases. We select transmitter 0.5. We attach antenna and leads, clean, dry, and connect a battery to one of them. Still no switch noise. We solder a 1000 mA-hr batter onto the transmitter (we want to make our first A3019E). No switch noise. We put electrodes in water, no noise. We remove transmitter 4.5 from water and short its inputs. We see the same 4-μV switch noise as when it is in water.

[06-MAY-11] Switch noise appears to have subsided substantially for the fourteen transmitters in water, as you can see here. The frequency of the 5.10 peak has moved from 18.625 Hz to 18.5 Hz. We compare the input noise to baseline EEG above.

We encapsulate our 0.5 transmitter. While the epoxy is still wet, we look for switch noise and find it. See here. We see amplitude 1.2 μV at 15.625 Hz. We put the transmitter in the oven to cure the epoxy. When we take it out, we cannot see the 15.625 Hz noise. We see no noise peak between 10 Hz and 25 Hz.

We take the fourteen transmitters out of their beaker of clean water, one after the other, and touch each with a magnet. After we touch the fourteenth, they are all turned off. We reverse the process, and all fourteen are on at the end. We immerse them all in warm 1% saltwater. We look at noise in 15-20 Hz. The noise peaks have vanished. Three hours later, the water has cooled to room temperature, and the peaks have returned.

[08-MAY-11] We placed the fourteen transmitters in pure water at 40°C. We lowered the temperature in stages with the help of water at 14°C from a drinking fountain. At each water temperature, we wait for one minute and measure the amplitude of the 18.5-Hz peak generated by 5.10's Hall Effect Switch. We obtain the following scary graph of noise versus temperature.

Figure: Switching Noise Amplitude on the Input of 5.10 versus Temperature. We determine the amplitude by looking at the peak in the 15-20 Hz band for an eight-second interval.

The graph gives the impression that the switching noise is going to keep going up sharply. But we doubt that is the case. We cooled our epoxy-encapsulated 0.5 transmitter with freezer spray and saw its switching noise rise up to a 2 μV. We cooled an unencapsulated transmitter and saw no switching noise. So far it appears that switching noise requires epoxy encapsulation and temperature below 25°C.

[09-MAY-11] We use the HES Acquire script with the Acquisifier to record temperature and switch noise amplitude in the 15-20 Hz band.

Figure: Switching Noise Amplitude versus Temperature. We present the amplitude for three transmitters. The amplitude is the rms value of the signal after filtering a four-second interval to 15-20 Hz.

We see that switch noise for 5.10 decreases below 15°C. The 15-20 Hz amplitude drops below 4 μV for all transmittesr at 30°C.

[11-MAY-11] We complete our first A3019E today. We used rejected circuit 0.5, whose current consumption is 120 μA (30 μA higher than we allow). We set it running to see how long it will last. In theory, it should run continuously for a year.

[01-AUG-11] Our A3019E has failed. There was a problem with the circuit to start with, which caused its excessive current consumption.

Failures

[17-APR-11] To date, we have shipped 3 A3019As to Oxford University, 27 A3019Ds to University College London, and 6 A3019Ds to Children's Hospital Boston. Of these, four appear to have failed. One at CHB (4.2) was hard to turn on after a week implanted, and impossible to turn on after two weeks. Another at CHB (4.1) generated its own 18-Hz noise after four weeks implanted. On 16-APR-11 we heard from UCL that two of their transmitters will not turn on. These were encapsulated on 02-DEC-11 or 08-DEC-11, and have taken part in at least two implantations each.

The following figure shows the noise we observe on CHB's 4.1 transmitter in a recording made at CHB on 8th April. For comparison, we show the simultaneous recording from 4.8. Both transmitters are implanted for EEG recording.

Figure: Voltage and Spectrum of Noisy Transmitter (4.1, Green) and Normal Transmitter (4.8, Blue). Taken from archive M1302289879. The peak at 18 Hz goes up to 300 counts, which is an amplitude of 150 μV. The noise on 4.1 extends up to 250 Hz, with some reduction at 160 Hz.

The Hall Effect Switch, U1 in the schematic, operates at a switching frequency of around 18 Hz, as we describe here. This switching noise is isolated from the transmitter power supplies by R1 and C1, and further removed by C2 and C3.

[18-APR-11] We hear from UCL that several more transmitters have failed over the weekend. Twenty weeks have passed since we encapsulated some of them, and nineteen since we encapsulated others. The average operating life of an A3019D is 18 weeks. If they were left on when they were stored, their batteries would be exhausted by now. But UCL assures us that the transmitters were off while set aside.

[22-APR-11] Here is a curious failure: full-range square wave at 1.2 Hz continuously for ten days, after six weeks running perfectly in a live animal.

Figure: Square Wave Failure on 3.8. Taken from archive M1302907845. At the time of the recording, No8 had been implanted for seven weeks. The problem started after six weeks implanted.

We inspected 4.1 and 4.2 from CHB today, and found no sign of damage to the encapsulation or the lead bases. The No2 antenna had come out of its tube, but it did not look corroded. The No1 antenna tube was unsealed at the tip, and there was corrosion on the length of the antenna, but none at the base, where the epoxy bead stopped further penetration.

[25-APR-11] Delayed failures can be caused by broken capacitors and resistors. Both can fracture during surface-mount assembly or when the board is flexed during subsequent handling. According to reports like this one, modern assembly methods cause very few fractures, so that most fractures these days are flex fractions.

We observed fractures in three resistors while testing 80 A3019 assemblies. On three other boards, C4 was incorrectly soldered. When we replaced C4, the replacement worked. Today we replaced C4 on a board and tested the original part afterwards. Its capacitance was 3.7 μF. This capacitor is hard to reach because it is tucked between three larger components.

The circuit board is only 0.8 mm thick, instead of the usual 1.6 mm. It is four times more prone to flexing than a standard board. The A3019 assembly is made in panels of ten, out of which the individual circuits must be broken along score lines. Further flexing could occur when we calibrate and encapsulate the sensors.

We have seen ceramic capacitor fractures before. A fracture might cause no problem for months. When a problem first appears, it is intermittent and temperature-dependent. Later, it becomes permanent. In our experience, a fractures result in an open circuit, although we read that capacitors can fracture and cause a short circuit. It could be that the oscillation we see above is due to a fracture in gain resistor R11. It could be that a failure of decoupling capacitor C2 will allow 18 Hz noise onto the amplifier power supply.

We have an A3019P that we dedicated to circuit tests. We left the inputs open circuit with no leads. We removed and replaced various components. We measured the input noise as we went along.

Several of the capacitors are redundant. Resistor R2 gives us a slight reduction in high-frequency noise from the digital circuits. We see no intrusion of 18 Hz noise when we remove C2 and we see no 1-Hz self-induced oscillations when we remove R11.

Another source of gradual failure is corrosion and by residual solder flux. We use water-soluble acid flux to attach the antenna and leads. We take care to soak the circuits in hot water, clean them with a brush, and rinse them in hot running water. We always inspect for flux residue. We note that R11 is next to the X terminals, and most certainly gets covered in acid flux while we solder the leads in place.

[26-APR-11] While making transmitters yesterday, we discovered that blowing air on our transmitters and letting them sit for an hour does not remove all moisture from beneath the components on the circuit. Perhaps it is trapped moisture that is causing our gradual failures. Trapped moisture eventually corrodes solder and copper joints, creating ionic compounds that conduct electricity. We changed our encapsulation procedure to include a one-hour back at 100°C to ensure that all moisture is evaporated before we encapsulate.

We receive recordings from Joost at ION from a set of 12 transmitters we shipped them at the beginning of January. Here is what we deduce from the recordings.

We damaged transmitter 5.4 today while clipping off the programming extension. We were holding it between thumb and finger. Later, we found that the transmitter would not turn off. We noted that U2, a tiny 5-ball BGA, was not level on its footprint, even though the balls had flowed. We removed U2 and found that the pad for U2-3 had snapped off. We loaded a new part and the transmitter works.

There were three A3019P circuits in our "failed circuits" bag. One of them would not turn on and off. We replaced U2. Now it works fine.

[29-APR-11] Today we turn on the noisy transmitter 4.1 we describe above, and find that its input is now showing full-scale oscillations at around 1 Hz, just like UCL's 2.9 and 3.8. The 4.1 transmitter X leads are soldered together.

Figure: Start of Square-Wave Failure. This transmitter, 4.1, previously showed 18-Hz noise (see here).

Note: On 04-MAY-11 we discovered that transmitter 4.1's X leads were both broken at the base, and we are able to make contact by flexing the leads.

[02-MAY-11] Transmitter 4.1 was off on Monday morning. We assume it turned off of its own accord over the weekend. We turned it on. The square wave was gone. The input noise with leads shorted together was only 8 &muV;, of which 4 μV was 18 Hz. With leads open, we see mains hum of only 40 μV and the sensor cannot detect heartbeat. We conclude that the gain of the amplifiers has dropped by a factor of ten. The transmitter may have turned off because its battery was drained.

[03-MAY-11] Transmitter 4.1's battery voltage appears to be down to 2.2 V today. We estimate battery voltage with 1.8×a÷65535, where a is the average value of X. We turn 4.1 off for four hours. When we turn it back on, its voltage appears to be 2.7 V again. We leave it running. The X input is much as it was yesterday. We note that 4.1 was encapsulated on 01-FEB-11 and shipped a few days later. That's thirteen weeks ago, which is not enough for the battery to run down in a functioning transmitter, even if it were left on all the time.

[04-MAY-11] We took a closer look at 4.1's leads and antenna. It turns out that both the X springs are broken near the base, as you can see in this photograph. The failure of this transmitter during implantation was an absence of EEG signal. In the absence of EEG, we noticed the 18 Hz noise. But the noise itself was not the cause of failure.

Battery voltage appears to be 2.0 V. If we press on the leads at the base, we can get one or both of them to make contact again, and at that point we can pick up mains hum, but sensitivity is still poor. We still see 4 μV of 18 Hz noise. We turn off and heat the transmitter for a minute, turn on again, and battery voltage rises to 2.6 V.

[09-MAY-11] We receive a detailed report from Rob at ION, describing the state of each of the 14 transmitters 3.1 to 3.14 we sent to him in January. One transmitter failed six weeks into its first implant, generating 1 Hz oscillations. Another failed six weeks into its second implant, generating noise. Another failed three weeks into its second implant, with intermittent connection to EEG. The others are working fine. One is a week into its third implant. None have exhausted their batteries.

[27-MAY-11] Three more implanted transmitters have failed at CHB. The 4.5 transmitter, which ran well for two weeks in its first implant, would not turn on after explanting. The 4.7 transmitter is not picking up EEG. The 4.6 transmitter suddenly started generating its own noise 17 hours after implantation, at time 3098 s in archive M1306224760. We see large transients for ten seconds, and then the signal shown below.

Figure: Noise on 4.6 and No Signal on 4.7.

In a letter to CHB, we propose that their antennas are penetrating the antenna tubes after being stretched to double their length during implantation and explantation. We place an old A3019A in water with a bare antenna. Its electrode leads are cut off at the base. We see noise of 1.8 mV rms, spread throughout the 160-Hz bandwidth as you can see here. When we place the antenna in water, but the leads out of the water, or visa versa, there is no noise. If we connec the analog leads with the water through our finger, while immersing the antenna in water, the noise appears. But at the same time, other transmitters immersed in the same water do not pick up the noise. The transmitter's own inputs are not differential with respect to the antenna noise, while those of other transmitters are differential.

Figure: Antenna Loop After Stretching to Double Length. For closeup of breach, see here.

The above photograph shows how an antenna will penetrate its insulating tube after being stretched by a 500-g force. In an encapsulated transmitter, this force is carried by the base of the electrode leads. The silicone around the bases is thick and strong, but the springs themselves can stretch within the silicone, and so become weakened or break. We suspect this is how the leads of transmitter 4.1 were broken.

[22-JUN-11] We receive from ION transmitters 2.6, 2.7, and 2.9. Transmitter 2.9 turns on and picks up mains hum with the correct gain. When immersed in water produces a white noise spectrum with a 2-μV peak at 20 Hz for its switch noise. We see no damage to the encapsulation or leads. Transmitters 2.6 and 2.7 will not turn on.

[15-JUL-11] After a few weeks in water, 2.9 picks up heartbeat but not mains hum. We see water beneath the silicone at the base of the leads. By squeezing in the right place, we see water emerging from between the leads and the antenna tip, where they are bound together by heat shrink. This transmitter is of an early construction, in which the binding is close to the transmitter body, creating a rigid lump. We moved the binding away from the transmitter later, and now we see that this is a good thing for two reasons. First, it relieves stress, and second it prevents the gap between the wire bundle from extending into the encapsulation. We play around with 2.9 some more. It is hard to turn on and off, and we observe the 1-Hz square wave that originally appeared during implantation. We put the transmitter in the oven and leave it to dry out.