Subcutaneous Transmitter (A3013)

Contents

Description

The Subcutaneous Transmitter (A3013) transmits biometric signals from within the body of a live animal. The A3013 detects biometric signals with thin wires and transmits them with another wire acting as an antenna. A nearby antenna receives the radio-frequency transmissions. The A3013 maintains robust data transmission from the body of a freely-moving rat to a receiving antenna at ranges up to 50 cm in all laboratories, and up to 300 cm in some laboratories, depending upon local radio-frequency interference and the prevelance of reflecting metal cabinets and tables. The A3013A will transmit 512 16-bit samples per second continuously for six weeks. We can turn the transmitter on and off at any time with a magnet, even when it is implanted in an animal. The A3013's shelf life is nine months. Its volume when encapsulated in epoxy and silicone is 2.8 ml. The encapsulated transmitter is small enough to implant easily in a rat.

Figure: Encapsulated Subcutaneous Transmitters A3013A-E. The encapsulation consists of an epoxy center with a silicone coating. The wires are all 125-μm 316-stainless steel with teflon insulation. The antenna wire is formed into a loop. The input wires are 150 mm long, and tinned at the other end using acid flux. The labels on the transmitters give the date (ddmmyy), the transmitter channel number (id), and the part number.

The A3013A has one two-wire, differential 10-MΩ input. Its dynamic range is 9 mV. A one-pole high-pass filter and a three-pole low-pass filter provide a bandwidth of 0.2 Hz to 160 Hz. Within this bandwidth the electrical noise on the input is 12 μV, which is 80 sixteen-bit ADC counts.

The A3013's radio-frequency transmissions occupy most of the bandwidth available in the 915-MHz ISM Band. Multiple A3013's operating in the same area can share the same receiving antenna. The A3013 radio-frequency transmissions are so brief that collisions between transmitters are rare. When collisions do occur, the data acquisition system eliminates the corrupted measurements from the data stream. The resulting degradation in signal quality is negligable. Ten A3013s can share the same receiving antenna and all maintain a signal bandwidth of 160 Hz simultaneously. If we drop the signal bandwidth to 40 Hz, forty transmitters could share the same antenna.

The A3013 is available encapsulated (-E suffix), naked (-N suffix), and programmable (-P). The encapsulation will endure freezing temperatures and vacuum without losing its integrity. It will keep water out of the transmitter for the duration of its operating life in a live animal. Such rugged encapsulation is hard to achieve. We describe the development of our encapsulation process in our SCT Encapsulation report. The stainless steel wires are designed to endure repetetive stress in the body of a live animal for several months. We describe our search for a flexible, fatigue-resistant wire in our Flexible Wires report.

Figure: Naked Subcutaneous Transmitters (A3013A-N), Top and Bottom Sides. The A3013A is a single-channel transmitter with three external connections: X+ is the positive input for channeld X, X− is the negative input, and RF is the antenna. Marked are (1) battery, (2) magnetic reed switch, (3) programmable logic chip, (4) 100-nF AC input coupling capacitor, (5) 10 MΩ input resistance, (6) analog amplifier, (7) the Y+ connection that is used only in two-channel circuits, and (8) the Y− connection not used in finished transmitters.

All versions of the A3013 start off with a programming extension that allows us to program the transmitter's logic chip, calibrate its radio frequencies, and set its channel number and sample rate.

Figure: Subcutaneous Transmitter with Programming Extension (A3013A-P).

The A3013 provides one or two inputs, depending upon how it has been programmed. On the A3013A and A3013B, channel X is fully-differential with dynamic range 9 mV, input impedance 10 MΩ, bandwidth 0.2 Hz to 160 Hz, and noise 12 μV rms. Channel X requires only two connections for a fully-differential measurement because the A3013 is battery-powered, which means its ground potential is floating. On the A3013B, channel Y is single-ended with dynamic range 54 mV, input impedance 10 MΩ, bandwidth 0.2 Hz to 20 Hz, and noise 25 μV. The Y channel has two connections, but we use only the Y+ connection in a live animal. The A3013B records the potential of Y+ compared to X−.

Some dual-channel transmitters alternate the samples they transmit between the two channels. The A3013C transmits both channels at 64 SPS, for a combined transmission rate of 128 SPS. Other dual-channel transmitters skip an occasional sample from X for a sample of Y. The A3013B, skips every eighth sample X for a sample of Y. Channel X receives 448 SPS, with the samples un-equally spaced, while channel Y receives 64 SPS equally-spaced samples. Dual-channel transmitters give each channel its own channel number in the messages they transmit. The receiver will record the two channels as if they were transmitted from two separate transmitters.

The sample rate and battery life of an A3013 varies from one version to another. The A3013A, for example, transmits 512 samples per second from a single channel, and will operate continuously for at least two months. The A3013C, on the other hand, transmits 64 samples per second from each of two channels, and will operate continuously for five months. When de-activated, all versions of the A3013 have battery life nine months. As you use up this nine months, the operating life is used up in proportion. After four and a half months on the shelf, an A3013A will have operating life four and a half weeks.

We receive A3013 transmissions with our Data Receiver (A3018). The Data Receiver stores the signal it receives from one or more Subcutaneous Transmitters. Each transmitter uses a different channel number to distinguish its transmissions from those of other transmitters. The transmitters share the same radio frequency band by each transmitting for only a tiny fraction of the time. We read out and display the transmitter signals with our Recorder Instrument. We analyze and store the signals with our Neuroarchiver Tool.

The A3013 is an improved version of the Subcutaneous Transmitter (A3009). See the A3009 manual for most of our work on the core circuits of the A3013. In the passages below, you will fine an account of our development and testing of the A3013, along with graphs and tables of its performance and specifications.

Design

Note: All our schematics and Gerber files are distributed for free under the GNU General Public License.

S3013_1: Logic chip, 900-MHz oscillator, battery, reference clock, antenna, analog inputs, anti-aliasing filters, sixteen-bit ADC, and amplifiers. Component values shown are for the A3013A.

S3013_2: Programming extension.

S3013X_1: The A3013X circuit, showing combined antenna and X+ connection with inductors acting as radio-frequency chokes. These inductors caused us problems.

P3013A: Firmware for logic chip of A3013A. Same firmware, with some constant changes, works in all other versions of the transmitter.

A3013A: Assembly spreadsheet for A3013A, including BOM, KIT, and COST.

A301301A: Printed circuit board files in zip archive.

Panel Drawing: Drawing of 6 × 2 panel.

Panel Files: Gerber files for 6 × 2 panel.

Versions

An A3013 part number has the form A3013f-p, where f is the transmitter function code and p is the transmitter packaging code. The function code specifies the behavior of the two input channels. The packaging code specifies the physical encapsulation or lack of encapsulation.

We set the function of a transmitter during assembly and programming. We change cacitors on the board to set filter frequencies. We change resistors to set the channel gains. We program the logic chip to set the sample frequency for each channel.

The A3013B transmits 512 messages per second, and shares these between two channels. It transmits seven messages in a row from X and then transmits one from Y. The result is 64 SPS for Y, and 448 SPS for X, but the X-samples are not regular. We can do this, and still maintain near-perfect 160-Hz bandwidth, because our data acquisition software will insert a substitute message in place of the missing message, and give this message the same sample value as the previous message it received from channel X. This substitution, combined with the sharp roll-off of our channel-X low-pass filter, give us near-perfect reconstruction of the original signal.

Once we clip off the programming extension, we can no longer modify the circuit's sampling frequencies. Once we encapsulate the circuit with epoxy and silicone, we cannot replace the battery. But we can replace the wires if we are willing to restore the silicone coating. Some users may like to attempt their own water-proof encapsulation. We recommend such users read our report on SCT Encapsulation before they attempt their own encapsulation. Other users may be planning to use the A3013 out of water, in which case it may make sense for them to receive the circuit with programming extension.

Prices of various versions when purchased from Open Source Instruments are here.

Features

The A3013 is similar to the A3009. It is compatible in its message transmissions. The A3013 incorporates the following enhancements.

The A3013 logic chip is housed in a 48-pin TQFP (thin quad flat-pack) package instead of a 56-pin BGA (ball grid array). At first glance, the TQFP uses three times the board area. It is 10-mm on each side instead of 6 mm. But the BGA requires so many vias that we cannot place parts on the opposite side of the printed circuit board. Therefore it occupies a 6-mm square area on both sides. Furthermore, the routing of signals to the BGA balls is so difficult that we cannot use the middle copper layer for any signals other than those destined for the BGA. It turns out that the BGA does not, in the final analysis, make the layout of the board any easier. If anything, the BGA may end up occupying more board space with its vias and tracks than does the TQFP. The advantage of the TQFP is that it is far easier to mount on the board, and there exist sockets that will allow us to program the TQFP before we mount it on the board. By programming before mounting, we can remove the programming connector from the A3013, thus saving more space, and allowing us to add more circuits (in this case, an extra analog input). We do not have to heat the board and the logic chip with a heat gun or in a toaster oven. We can mount the chip with a soldering iron. We believe that the lower-temperature mounting of the TQFP is responsible for the lower power consumption of the final circuit (70 μA for the A3013A instead of 90 μA for the A3009A).

The A3013 circuit supports two analog inputs. One provides a three-pole low pass filter and a gain of up to 1000. The other provides a one-pole low-pass filter and a gain of up to 100. The schematic shows gains of 300 and 50 respectively for these two channels.

The A3013 firmware incorporates some new code that supports transmission from two channels.

One enhancement we planned for the A3013 was the combination of the X+ and antenna leads. Our original circuit provided two inductors in series with the X input leads that would isolate them from the A3013 ground plane at RF frequencies. In theory, this isolation would the X inputs to carry RF power and radiate it into space, even as they carried the low-frequency analog inputs. But the combination proved to be inconvenient, in that we could not connect X+ to a low-impedance signal source during tests, and a failure, in that the inductors caused our RF oscillator to produce four times as much power for a little while until it failed. By removing L1 and replacing L2 with a 1-kΩ resistor, we were able to return to the three-wire solution we used in the A3009.

Programming Extension

The programming extension, which we show in the figure above, provides three connectors. These are shown in the second page of the schematic. Connector P1 is an eight-way single-row header. It is the logic chip's programming connector. Connector P2 is a two-way single-row header with a polarizing tab. It is the battery power connector. The circuit board shows you which pin is 0V and which is 3V. You can power the transmitter with a 3-V power supply, or you can plug in a battery with a cable-mounting socket, like the one shown below.

Figure: Battery with Socket for use with Programming Extension.

The third plug on the programming extension provides three test pins. One is connected to the transmitter's 1.8-V logic power supply. The other two are outputs from the logic chip. The current firmware makes TP1 the transmission bit stream, so you can use it to trigger an oscilloscope and watch signals in an RF receiver with another probe. The TP2 signal is undefined.

The transmitter with programming extension is convenient for all tests but one: the test of operating range. The signal traces joining the transmitter to its extension, and the connection from the extension to the battery or power supply, act as un-tuned antennas for the RF transmission. We get best performance from the programming extension when we hold the entire assembly by the battery. Even then, its operating range is no more than half that of a transmitter without the programming extension.

We conclude that the transmitter with its extension is useful for tests of the analog inputs, current consumption, and battery life. But it is not useful for measurement of operating range.

Transmission

The A3013 transmits analog signals by sampling them and transmitting the samples as radio-frequency messages using its quarter-wave antenna. Most of the time, the transmitter logic and transmission circuits are inactive. Only the analog and clock circuits are operating.

When it comes time for a sample, the logic and transmission circuits activate, select and digitize one of the analog inputs, and transmit it using the antenna. The message frequency is the number of times per second the transmitter activates and transmits a message. The bit rate is the rate at which the transmitter sends bits within a message. The A3013 bit rate is 5 MBPS (megabits per second). One bit takes 200 ns to transmit. The entire message, including start bits and checksum, takes 7 μs. The A3013 message rate depends upon the version. The A3013A and A3013B message rates are 512 SPS (samples per second). The A3013A samples one input channel at 512 SPS. The A3013B samples two input channels, one at 448 SPS, the other at 64 SPS.

The A3013 use the message format described here, which is the message format used by the A3009. Transmissions from the A3013 are compatible with those of the A3009, and you can receive them with the Data Receiver (A3010) or the Data Receiver (A3018).

The A3013 applies a 1-GHz carrier signal to its antenna and makes small changes in the carrier frequency to convey information. We say 1 GHz, but the transmitters can be programmed to operate anywhere between 875 MHz and 1050 MHz. All existing versions of the A3013 operate in the 902-928 MHz band.

The best length for the antenna is one quarter the wavelength of the radio-frequency signal it transmits. In air, with a 915-MHz carrier, this length is 80 mm. Beneath the skin of an animal, however, the best length is shorter because radio waves travel more slowely in the presence of water. After much experimentation with antennas, we settled upon the bent 80-mm antenna as the best choice for operation in all environments.

Our favored modulation depth is ±4 MHz with the A3013, but we operated most of our A3009 transmitters with ±8 MHz modulation. Because of its narrower modulation width, the A3013 spectrum is narrower: roughly 18 MHz wide compared to the A3009's 26 MHz. Nevertheless, the A3013 and A3009 transmission power and spectrum are identical for the same modulation width. We discuss the A3009 spectrum in detail here.

A narrower spectrum means that a reception dead spot caused by reception failure at a particular frequency has to lie within the narrower spectrum in order to corrupt a message, so our resistance to multi-path interference increases as the spectrum width decreases.

A counter on the transmitter counts periods of its reference clock, and determines the sample frequency. The reference clock runs at 32.768 kHz, ±20 ppm. We call the reference periods ticks. Each tick is 30.5176 μs. If we have two transmitters sharing the same receiving antenna, and each transmits with an exact and constant interval between samples, we will find that the transmission instants coincide every hour or so for a period of several seconds. During this time, the transmissions collide and destroy one another.

To avoid collisions, all versions of the A3013 use the lower four bits of the current ADC word as the source of a random number to displace the next transmission instant. Transmission takes place the next time the counter has value equal to the last four bits of the previous transmission's DAC word.

The earliest transmission instant is tick zero, and the latest is tick fifteen. The average instant is 7.5 ticks, and the range is ±7.5 ticks. If we trigger on the instant of one transmission, and display the next on the oscilloscope, the second transmission occurs an average of T ticks later, where T is the nominal sample period in ticks. In the case of our 512 SPS transmitters, T is 64. So the next transmission occurs on average 64 ticks later, but with a distribution of ±7±7, which is a binomial distribution with total width ±15, and a peak at zero.

Figure: Transmission Scatter. The transmitter's sample frequency is 512 Hz. The width of the distribution of sample separation is 924 μs, and samples are more often at the center of this width.

In the figure above, we see the scatter of the next sample relative to a first sample. The width of the scatter is 924 μs, which is 30 ticks, exactly as we expect. The mean sample separation is a little less than 2 ms, which is correct for 512 SPS.

As we mentioned above, a dual-channel transmitter sends samples from its two channels with separate IDs, so that the receiver stores the samples as if they come from two different transmitters. The X channel number of a dual-channel transmitter is always even, so it can be 2, 4,... 14. The Y channel number is always one greater than the X channel ID.

The simplest way to divide the transmitter between two channels is to alternate between channels, so that you transmit both with the same sample rate. That's how the A3013C works. But the A3013B does something different. The bandwidth of X is much greater than that of Y, so there's no point in transmitting Y with the same sample rate as X. Instead, the A3013B transmits 7 samples of X at 512 SPS, and then skips an X sample to transmit a Y sample. Because the data recording software will be expecting an X sample at this instant, but will fail to find one, it will insert a substitute message whose value is equal to the previously-received sample. Most of the time, the substitute message does a good job of recovering the original signal, and when it does not do a good job, the failure is not dramatic.

Frequency Selection

The A3013 firmware configures the circuit's logic chip. The logic chip controls both the radio frequency and bit rate of the messages transmitted by the A3013. Resistors R4 to R8 form a 5-bit DAC (see schematic). This DAC drives the TUNE input of the voltage-controlled radio-frequency oscillator, U3. For a graph of radio-frequency versus TUNE voltage for this chip, see here.

Two parameters set the radio frequencies used by the A3013.

frequency_low=10; "low frequency."
frequency_step=2; "high frequency - low frequency"

The frequency_low gives the A3013 5-bit DAC value at which its MAX2624 transmits its logic zero (LO) frequency. The A3013 adds frequency_step to frequency_low to obtain its logic one (HI) frequency. When frequency_setp is 2, the LO frequency is 8 MHz above the HI frequency. When we calibrate an A3013, we adjust frequency_low until the LO and HI frequencies are eqully spaced around 915 MHz at 20 °C. Frequency variation between chips is ±10 MHz, even within the same batch of chips. So each transmitter must be calibrated. For a more detailed description of the calibration process, see here.

Two more parameters determine the bit rate.

tcd_divisor=22; "divide ring oscillator to get TCK"
half_tcd_divisor=14; "determines mark-space ratio of TCK"

During transmission, the logic chip runs a ring oscillator. The frequency of the ring oscillator varies from one chip to another, although chips within the same tray are almost identical. We divide the ring oscillator frequency to obtain the bit frequency. So we have to calibrate the bit frequency of each transmitter also, or at least check that it is correct.

Because we must calibrate the logic chip's ring oscillator, and the radio-frequency oscillator on each transmitter, we must be able to adjust the firmware on the logic chip after the circuit is assembled. This calibration takes place by means of the transmitter's programming extension. The calibration extension holds a programming connector, a power plug, and several test points. In the first A3013s we made, the calibration extention was a border around the transmitter footprint, which we must cut off with shears.

Current Consumption

The current consumption of the A3013A while active and transmitting its messages at 5 MBPS is ≈80 μA. While inactive, the current consumption drops to ≈20 μA.

The active current consumption of the A3009 was higher by roughly 15 μ. The A3009 had many problems with erratic current consumption as a result of our over-heating its BGA logic chip during assembly. We have observed no such effects in the A3013, nor do we expect to. The A3013 uses a TQFP package instead of a BGA, and we have long experience soldering TQFP packages. For more current consumption measurements see below.

The ADC on an A3013A broke. This was while we were experimenting with 2.5 MBPS messages. The message duration increases from 7 μs to 12 μs when we drop from our usual 5 MBPS down to 2.5 MBPS, so the message-transmission component of the active current consumption almost doubled. While we pulled parts off the board trying to find out which one was broken, we measured the current consumption of the circuit in various states.

As you can see, the inactive current remains unchanged at the lower bit rate, but the active current increases by roughly 35%. The battery life would therefore be 35% shorter. See here for a comparison of the two bit rates.

From the above results, we conclude that the current consumption of an A3013 transmitting at bit rate 5 MBPS is roughly 20 μA + 120 nA/SPS. AT 512 SPS, that's 80 μA, and at 128 SPS, it's 35 μA.

Battery Life

Here we present the results of our laboratory measurements of operating life and shelf life. For measurements in animals, see Live Animals.

The A3009 had a current consumption of 90 μA and a battery life of eight weeks (as we showed here). The A3013A's current consumption is around 80 μA, so we expect its battery life to be nine weeks. But we have not yet allowed an A3013A to operate continuously and run down its battery. The shelf life of the A3013 we expect to be around nine months.

If we assume a battery capacity of 120 mA-hr, and use the formula for current consumption we arrived at above, we expect a transmitter's battery life to be 1500 hrs (9 weeks) at 512 SPS for both channels combined, and 3500 hours (20 weeks) for 128 SPS.

[03-MAR-08] For the past few weeks we have been running an A3013A, channel number 5, from a near-exhausted battery. Today the average X input is 52300, indicating a battery voltage of 2.26 V (65535 counts is VBAT and the average of X is 1.8 V). We measure the battery voltage directly with a voltmeter and found it to be 2.27 V. This tells us that the average value of X is a good measurement of VBAT even as we near failure, and that the circuit works down to at least 2.27 V. (Raw data here.)

[05-MAR-08] Transmitter is still working. Reception is robust at range 100 cm. Average X is 56533, indicating VBAT = 2.09 V. (Raw data here.) We observe reception of 60-Hz hum, and step changes when we touch the electrodes. (Raw data here.)

[11-MAR-08] Transmitter 5 has stopped transmitting. Battery voltage is 1.6 V. We give it a new battery of a new type: the CR2032, which is 20 mm in diameter and has a capacity of 225 mA-hr. Our usual battery is the BF1632, a 16-mm diameter battery with 120 mA-hr capacity. We start recording once per hour to see how long the battery will last. This A3013A is one of our rejects. It's quiescent current is 95 μA (15 μA higher than our expected 80 μA). We expect the A3013A to run continuously for 225 mA-hr ÷ 95 μA = 2400 hrs.

Figure: VBAT versus Time with 225 mA-hr Battery. The battery is a CR2032. The transmitter message rate is 512 messages per second. We measure VBAT by taking the average of one second's worth of X samples. The average value of X corresponds to VCOM, which is 1.8 V as set by a regulator in the transmitter circuit. From this we deduce VBAT, which corresponds to a X = 65535.

[24-APR-08] The transmitter has been running continuously for 1000 hours. We see the battery voltage drop from 2.98 V to 2.86 V in the first few hours of operation. Since then the battery voltage has climbed slowly to 2.90 V. We reduce our recording rate to once per day.

[13-JUN-08] At 2300 hrs, battery voltage has dropped to 2.4 V. Our expected battery life is 2400 hrs. The transmitter is still functioning.

[25-JUN-08] The transmitter failed at 2425 hrs. We are now confident that we understand how to relate battery capacity to transmitter operating life.

[03-SEP-08] We took out a A3013D-N transmitter (two channels with 256 samples per second) that received a new battery on 27-DEC-07. Nine months have passed since the new battery was installed, and for that time the transmitter has been inactive. We turned it on and received 100% reliable transmission with no transmitting antenna when the transmitter was 5 cm from the receiving antenna. The channel noise was 100 counts on X and 20 counts on Y. The average value was 47871 on X and 46312 on Y. From this we deduce that the battery voltage is 2.47 V, which means the battery is 90% exhausted. We see that the shelf life of an A3013 is roughly 10 months, with a 10% loss of battery life every month.

[01-OCT-08] The A3013D-N still works when we turn it on. The average value was 60500 on X and 57700 on Y, suggesting a battery voltage of 1.95 V. With a voltmeter we measure 2.08 V.

Analog Inputs

The ADC (U5 in the schematic, an LTC1864) converts either its X or Y input into a 16-bit number. A 16-bit count of 0 means an input voltage of 0V. A count of 65535 means a voltage equal to the 3VA supply, which starts off at about 2.7 V when the battery is fresh, and drops to about 2.2 V just before the transmitter fails. As you can see from this graph, the battery voltage drops quickly in the last few days of the transmitters life, but up until that time remains approximately steady at 2.7 V. We assume a battery voltage of 2.7 V when we calculate the analog input dynamic range, but we do not claim that the absolute values returned by the transmitter are an accurate measure of the absolute applied voltage on a time-scale of days or weeks.

The dynamic range of either input channel is 2.7 V divided by the channel gain. In the case of channel X in the A3013A or A3013B, the dynamic range is 2.7 V / 300 = 9 mV. Each ADC count represents a 140-nV change in X. The dynamic range of Y in the A3013B is 2.7 V / 50 = 54 mV (the A3013A does not provide a Y channel). Each ADC count represents an 820-nV change in Y.

When we short the inputs of an A3013A together, we observe noise of standard deviation 70 ADC counts. This 70 ADC represents is 10 μV rms noise at the X input. If we remove R12 in the schematic, we see noise of 12 ADC counts, which means the 10 μV is indeed arising at the X input. The thermal noise generated by a 10-MΩ resistor in the 160-Hz pass-band of the A3013A is 5 μV. Our input noise is twice the theoretical minimum possible with a 10-MΩ signal source. With this we are well-satisfied.

But it is clear that our input noise is not generated by resistors. The inputs are shorted together, and the only remaining resistor on the input of the amplifier (U6 in the schematic) is R10, 200 kΩ. The thermal noise at the input should be less than 1 μV. What we are seeing in our 70 counts of input noise is the noise generated by U6, an OPA2349. The OPA2349 data sheet specifies a typical input noise of 8 μV p-p noise in the 0.1 Hz to 10 Hz range (that's pink noise, or 1/f noise), and 300 nV/√Hz above 10 Hz (that's white noise). In terms of ADC counts, we can expect 60-count fluctuations from one second to the next, and 30 counts of white noise. If we look at the input noise, we do see fluctuations combined with white noise, and the combination amounts to anything from 60 counts in one transmitter, to 100 counts in another. We also observe the size of the noise to vary with time in a single transmitter.

We conclude that our circuit-generated noise is dominated by that of our input amplifier, and that this noise can vary unpredictably with time, and from one transmitter to the next. So far, we have observed input noise varying from 60 to 100 counts when measured over a one-second period.

With 0 V applied to either input, the nomial voltage applied to the ADC input will be 1.8 V, which is VCOM in the schematic. Offsets in the amplifier op-amps can be as large as 10 mV for each stage, so the actual zero-signal voltage applied to the ADC might be higher or lower than 1.8 V, depending upon the final-stage gain. With a gain of 50 in the final stage, as applies for Y in the A3013C, the zero-signal voltage could be as high as 2.3 V and as low as 1.3 V. Because of this, the dynamic range with respect to the zero-signal can be anything from −46mV..+8mV to −26mV..+28mV. For channel X the problem is less severe, because the final-stage gain is 11, not 50. The range can be anything from −5.6mV..+3.3mV to −6.3mV..+2.6mV.

We connected the X input of an A3013X to our function generator through two 10-MΩ resistors. On resistor was in series with each of the ±X inputs, and we obtained attenuition of the function generator output by connecting the ±X inputs with a 1-kΩ resistor. The voltage applied to the X was therefore the function generator voltage divided by 20,000. We applied a 10-Hz sinusoid of varying amplitude to the X input using this arrangement of resistors, and obtained the following results.

We applied a 10-Hz, 5-μV amplitude input to X and recorded the transmitter for four seconds. The following figure shows the fourier spectrum of the output.

Figure: Spectrum with 10-Hz 5-μV Input. The frequency divisions are 10 Hz. The spectrum is scaled to fit the vertical axis. Note the sharp 10-Hz and 60-Hz components.

As you can see from the figure, the 10-Hz, 5-μV signal stands out in the fourier spectrum, above the white background noise and mains-frequency (60 Hz) hum. If, however, we gather fewer data points, our 10-Hz signal does not stand out as much. Here is 800 ms of data with a 10-Hz, 10-μV input.

Figure: Signal (left) and Spectrum (right) with 10-Hz 10-μV Input. The frequency divisions are 10 Hz. The spectrum is scaled to fit the vertical axis.

The 10-Hz, 10-μV signal is almost swamped by noise in the signal trace. In the spectrum, the 10-Hz stands out, but is only a few times more powerful than the noise. We conclude that the transmitters sensitivity to a signal depends upon the spectrum of the signal itself and the amount of time for which we can observe it. Given four seconds, the A3013 can detect a 2-μV, 10-Hz signal. Given only 1 second, it can detect only a 10-μV, 10-Hz signal.

We taped the bare ends of X+ and X− of an A3013B to the chest of a female volunteer. We obtained the trace on the left in the figure below. Our volunteer run up and down the laboratory corridor a few times, and did some jumping jacks. We obtained the trace on the rigth. We conclude that the short spikey features in the recorded signal are her heartbeats.

Figure: Heartbeat Before Exertion (left) and After Exertion (Right). Blue trace is signal from two bare wires taped 50 mm apart on volunteer's chest. Red trace is clock channel. The entire trace occupies 3 seconds, during which we see three heartbeats on the left and five heartbeats on the right.

The amplitude of the heartbeat signal is roughly 1000 ADC counts, or 150 μV. In the left trace, our volunteer took slow, shallow breaths, and stood still. In the right trace, she breathed more deeply, and stood less still. We took enough traces to conclude that the low-frequency rumble on the right-hand trace is due to her breathing and movements.

Reception from the transmitter while taped to the volunteer was less robust than when held in free air. We had to keep the transmitter within 1 m of the antenna, and in a favorable orientation, in order to obtain reliable reception. We wrapped tape around the end of X+ to insulate it, to see if connection of X+ to the body was the problem. But reception remained poor. If we move the transmitter itself away from the body, reception improves. The antenna is less efficient when it it lies on a large, flat, conducting body such as the human chest.

The three-pole low-pass filter we use on channel X in the A3013A has a cut-off frequency of 160 Hz. The filter is identical to that of the A3009, whose frequency response we show here. The three-pole low-pass filter we use on channel X in the A3013D has cut-off frequency 80 Hz. The one-pole low-pass filter we use on channel Y in the A3013D has cut-off frequency 20 Hz. We applied a 1-mV amplitude sine wave of varying frequency to X on an A3013A, and then simultaneously to X and Y on an A3013D. We recorded the rms amplitude of the transmitter output in units of ADC counts and obtained the graph shown below.

Figure: Frequency Response of Low-Pass Filters. 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 normalised with respect to the response at 0.5 Hz.

When it came to testing the analog inputs of our production circuits, we used the test apparatus shown below. At the time we performed our tests, all the circuits were A3013Xs, with a combined antenna and X+ wire.

Figure: Analog Input Test Apparatus. We apply a sine wave or square wave to the black and red clips. The small resistors are 1 kΩ. The large one is 10 MΩ. One of the 1-kΩ resistors terminates the red wire. The other is part of a divide-by-10,000 voltage divider.

As you can see, the red antenna wire is terminated by a 1-kΩ resistor. We cannot connect the red wire of an A3013X directly to the black or red clip without compromising radio-transmission (see below).

We connected each transmitter to the test apparatus as shown, and applied a 10-Hz, 20-V peak-to-peak square wave to the black and red clips. The signal applied to the transmitter inputs is 2 mV peak-to-peak. We checked that each transmitter produced a waveform like the one below.

Figure: A3013A Ouput for 10-Hz 2-mV Square Wave Input on X.

We also tried several other inputs on some of the transmitters, to make sure that they met our expectations with noise, gain, and frequency response. We stored the raw Recorder Instrument data for you to look at yourself. To view the raw data, download the file and open it from the Recorder Instrument in LWDAQ. You can look at Noise taken with a 1-kΩ resistor across the X inputs, 2-mV 10-Hz Square Wave, 200-uV 10-Hz Square Wave, and 40-uV 100-Hz Square Wave. If you open the Neuroarchiver, you can look at the spectrum of these waveforms also.

We recorded X overnight from an A2013A. The transmitter was coated with our Process V, immersed in salt-water, with the X wires soldered together. We monitored X in 5-second intervals and recored its average value in each interval as well as its standard deviation.

Figure: Average and Standard Deviation of X versus Time. We monitor X in 5-second intervals to calculate its average and standard deviation. The vertical scales are in sixteen-bit ADC counts, where 100 counts is 13 μV.

The average value of X corresponds to VCOM, or 1.8 V. The value of X in ADC counts will rise as the battery voltage, VBAT drops, because X = 65535 × VCOM / VBAT. In this case, X rises by roughly 300 counts in 17 hours, which corresponds to VBAT dropping by roughly 20 mV. The standard deviation of X corresponds to the analog input noise. The standard deviation is almost always 100 counts, which corresponds to 13 μV. But we see spikes in both the average and standard deviation of X. The following graph shows a close-up of such a spike.

Figure: Average and Standard Deviation of X versus Time. This graph is detail of a spike from the graph above.

Something forces a sudden and temporary jump of 65 μV in X during the night. The jump in X endures for roughly one minute, and then subsides. The standard deviation of X rises during the jump up and the jump down, but otherwise remains close to 100 counts. We do not know the cause of these spikes.

External Noise

For sources of noise inside the A3013 circuit, including thermal noise, see the the section above. Here we discuss external sources of noise. One external source of noise is mains hum. In the USA, mains hum is 60 Hz noise from power cables, and in Europe is 50 Hz noise from the same sources. To understand sources of mains hum, please consult our Introduction to Mains Hum.

We measured the mains hum received by channel X of our A3013X for various combinations of wires and resistors across its terminals over a metal optical table (where we found there was a lot of mains hum).

The mains hum cannot be induced by magnetic fields, or else the noise would be greatest when the resistance was smallest. The 10 MΩ input resistance of channel X forms a loop with the lead wires, and this loop would pick up and transwer most energy to the 10 MΩ input resistance when there was no other resistance in the loop to compete with it for voltage. We conclude, as we did here, that mains hum is coupled capacitively into X. The smaller the resistance across the inputs, the greater the load upon the capacitive impedance that supplies the mains hum, and the smaller the mains hum.

Figure: Transmitter (A3013X) with 80-mm Twisted Leads. Left side standing on desk with 1 MΩ Load. Right side being wiggled with no load resistor. Note that this transmitter has a combined antenna and X+ wire, which we later abandoned because of complications.

We now took the same transmitter, with 80-mm twisted leads, and placed it over a desk instead of the metal optical table, where there was less mains hum. We measured the mains hum with the transmitter with various loads. We rotated the transmitter about the vertical axis. In each orientation, the wires were straight up, as they are in the figure above, and the transmitter board remained centered above the same spot on the desk.

Orientation does not appear to affect the amplitude of the mains hum. Holding our hands near the transmitter, and wiggling the transmitter, cause large signals on the input, but these are not mains hum signals. Here is what we see when we wiggle the transmitter and wave our hands near it.

Figure: Interference. Left is while wiggling the transmitter with no load resistor. Right is while moving hands near input leads. Vertical scale is 1 mV/div, horizontal extent it 0.8 s.

Wiggling the transmitter creates a large signal on X. This signal vanishes when we connect the ends of the input leads with a 1-kΩ resistor. If the signal were caused by magnetic induction, the signal would remain even with a 1-kΩ resistor. Therefore what we are seeing is the movement of the two electrodes through an electrostatic field above the desk. When we move our hands near the electrodes, we create or disturb this field, and so cause interference.

When we leave the transmitter stationary, we see no signal from the electrostatic field because our signal passes immediately through a passive high-pass filter with time constant 100 ms. But we predict that there will be a potential difference between the two electrodes due to the prevailing electrostatic field.

We plan to perform experiments with charged insulting objects moving near open electrodes to confirm our hypothesis. If the source of these signals is indeed electrostatic fields, then we can understand the Delta Power we observed in our mouse earlier. When the mouse moved the electrode wires, which were external to its body, around in the electrostatic field created by rubbing its fur against the cage wall, it generated interference on the electrodes in the 1.5 Hz to 6 Hz band.

If we want to avoid electrostatic noise on our electrodes within an animal, we must make sure that they are enclosed completely by the animals skin. The skin is not a perfect conductor, but we suspect that it conducts well enough to remove electrostatic fields in the region of the electrodes.

Encapsulation

We discuss our development of an encapsulation procedure in SCT Encapsulation. Our work is still on-going. At the moment, our favored method is to fill the gap under the battery with silicone caulk and coat four times by dipping in silicone dispersion. We also fill the short piece of heat shrink tube that bundles the wires with caulk before we heat it up, so the space inside the heat shrink is water-tight. We place the heat shrink 10 mm from the transmitter body. The resulting coating is tough on the outside, allows us to wiggle the wires without breaking the coating. In a vacuum, there is so little air trapped within the coating that expanding bubbles do not burst.

Damaged Transmitters

In the A3013X, the X+ wire acts as the RF antenna, as well as the positive input to the X input channel. The red wire will acts as an effective antenna provided you terminate it with a contact resistance of at least 1 kΩ.

In order to share the same wire for both the low-frequency analog X+ input and the radio-frequency output, the A3013X circuits used a 100-nH inductor to isolate the analog input from the antenna wire. This inductor is L1 in the original schematic. Inductors on the outputs of radio-frequency amplifiers can cause oscillations, as we observed here. The first A3013X circuit we made performed well. Chip U3, the radio-frequency oscillator, came from the first of two tubes of MAX2624 chips. But transmitters made using chips from the second tube failed at random. Some of them stopped transmitting altogether. Others introduced noise into the X input.

Before failure, and after failure in the case of transmitters that produced noise, the operating range of these A3013X transmitters was twice as great as those of transmitters made from chips taken from the first tube. This suggests that the radio-frequency output power has quadrupled. We conclude that the MAX2624 output buffer is near to instability in the presence of L1. Chips from our first tube were stable and operated normally. Chips from the second tube were unstable. Their power output increased by a factor of four, which caused them damage.

Figure: Noisy Signal from Damaged Transmitter. This transmitter is coated in silicone dispersion and immersed in water. The red wire is connected to VBAT as a result of failure of U3. We leave the ends of the wires out of the water, or else conduction between the two will drain the battery. If we connect the two wires, or load them with a 1-kΩ resistor, transmission stops.

One solution to this problem was to use chips only from our second tube. But we decided to eliminate L1 and add a separate antenna wire to the transmitter. It occurred to us that we could fasten this wire to the X wires, and so keep its tip from drifting around inside the animal's body. With the X input wires free of the radio-frequency signal, we can connect them to low-impedance signal sources, which makes them easier to test.

The A3013A has a separate antenna wire, as you can see here. While studying the effect of antenna length and diameter (see below), we discovered that an active transmitter is vulnerable to contact between a metal tool and the antenna wire. In particular, cutting the antenna or soldering the antenna while the transmitter is active is likely to cause the RF power source to fail. We must make sure the transmitters are turned off while we work on their wires, and we make sure that we pull some of the antenna insulation over the tip of the conductor to protect it from contact with metal tables and tools.

Volume Measurement

The shape of our transmitters is approximately rectangular, but not rectangular enough for us to use their rectangular dimensions to obtain a good estimate of their volume. The figure below shows our apparatus for measuring transmitter volume.

Figure: Apparatus for Measuring Transmitter Volume.

We fill the blue cup with water until the surface is flat and flush with the rim of the cup, as seen when looking across the water surface from the side. We place the object whose volume we wish to measure in the cup. Provided that the object's volume is less than 5 ml, the water does not spill out, but bulges upwards with surface tension, as shown in the figure. Now we remove water with a syringe until the surface is once again flat. The volume of water we removed is the volume of the object we placed in the cup.

We calibrated the measurement with four half-inch steel spheres. These each have volume 1.07 ml. Our measurement of the volume of two such spheres was on average 2.1 ml, with precision 0.1 ml.

As you can see from the table, the volume of our silicone-encapsulated A3013A is 2.0 ml. Our Process X encapsulation adds 0.8 ml to the volume, but has the great advantage of being water-proof and vacuum-resistant.

Individual Characteristics

For a picture of our first twelve A3013 circuits, see here. Two of them are in our epoxy mold. These later failed after we broke them out of the mold.

We paint the serial number, version letter, and channel number of each transmitter on the bag in which we ship it to you. We do not mark the transmitters themselves. When the transmitter provides two channels, the number marked on the transmitter will always be even and correspond to channel X. Channel Y will be transmitted with an number one greater than X.

The table above shows how the current consumption of our first twelve production-version A3013s varied from one circuit to the next. The consistency we see in the current consumption suggests that our assembly process is in no way damaging the logic chip. The variation in the DAC value required for a 915-MHz low-frequency shows variation between oscillator chips.

Two transmitters suffered from roughly 10% missing messages. These were numbers 007 and 011. In August 2007, we refurbished 011 and measured its RF output power spectrum with our new A3008B spectrometer. The RF spectrum covered 924±9 MHz. We placed a 4-kΩ resistor across R4, and so dropped the spectrum to 917±9 MHz. The missing message rate for 011 after this modification dropped to 0%. We conclude that improper initial calibration of the Low Frequency ADC Count caused the missing message rate in both these transmitters.

Antenna Experiments

Here we present new measurements and observations made with the help of the A3013A.

Diameter

As we report in SCT Encapsulation, stranded wires conduct water by capillary action, and so transport bodily fluids into the transmitter, disturb its analog inputs, and drain its battery. Solid wires appear to be immune to this capillary action because they do not have spaces inside the insulation along which water can creep.

We are happy to use teflon-insulated, solid, silver wires for our X inputs. But we wonder if the antenna will be efficient with such a thin wire. We set tried various types of wire for our transmitting antenna, and recorded the power received by our RF Spectrometer (A3008B) at a range of 30 cm for each antenna. The transmitting antenna was vertical in each case. The receiving antenna is a Loop Antenna (A3015L) You will find our results below.

It appears that the diameter of the conductor plays very little part in its efficiency as an antenna. We see no reason why we cannot use 0.2-mmm solid silver wire as our antenna, as well as our X inputs. Even 0.1-mm copper wire performed well as an antenna.

Voltage

According to the MAX2624 data sheet, our RF output is coupled through a capacitor on the chip. But we find that we cannot connect a loop antenna directly to the RF output and ground the other end. The MAX2624 stops working when you make a DC connection between OUT and 0V. In our damaged transmitters, the damage had caused OUT to be connected to VBAT (see schematic). Among transmitters that have been operating for several weeks, roughly a third have OUT connected to VBAT. But freshly-made transmitters never have OUT connected VBAT.

We cannot make electrical contact with 0V when a transmitter is encapsulated, but we can get to VCOM (which is 1.80 V) at X−. We measured the voltage between OUT and VCOM in one of our encapsulated transmitters, and found it to be 0.95 V. This suggests OUT is connected to VBAT, and that VBAT is 2.65 V. When connect OUT and VCOM with 10 kΩ, the voltage between OUT and VCOM drops from 0.95 V to 0.2 V. We conclude that OUT is connected to VBAT with a DC resistance of order 40 kΩ. This connection was not present when we first turned the transmitter on. At some point, during weeks of operation, the MAX2624 suffered some kind of internal damage, and as a result, its OUT pin became connected to VBAT with a 40-kΩ reistance.

When we immerse a transmitter in water, it is possible for conduction to occur between the X− and antenna leads. If conduction occurs with impedance less than a 1 MΩ, we will find our battery life decreased. Future transmitters provide a discrete capacitor in series with the antenna, to make sure there is no DC current path between X− and either 0V or VBAT. But such a capacitor will not protect the transmitter from internal damage, as we showed in Damaged Transmitters.

Two encapsulated transmitters showed VBAT on the antenna. After working with both for a while, using different antenna lengths for the experiment described below, both transmitters stopped working. Several months later, we received three transmitters back from London. They all worked when we first turned them on, but their antenna wires were broken off. When we soldered new wires onto the transmitter contact pins, we connected the three pins together with a large solder ball to clean them. After this experience, two out of the three transmitters were broken.

Length

The encapsulated transmitter shown here has its output leads soldered to pins. It is our first water-proof tranmsitter with output pins, and our first opportunity to vary antenna length in water and measure transmitted power. Once we are finished cutting back the antenna, we can replace it with a new wire.

We placed a transmitter in a 2-liter beaker of water by holding it by the body just below the water surface in the center of the body of water, with a straight antenna facing down. We placed the beaker 50 cm from a Loop Antenna (A3015L) and conneted the antenna to a Spectrometer (A3008B). We used our LWDAQ Spectrometer Tool to measure the power received by the antenna.

Our antenna is a 0.28-mm diameter teflon-insulated wire for our antenna. We started with a total antenna length of 125 mm, including the 4-mm connector pin, which protruded from the transmitter's ground plane, but not including the 5-mm trace from the RF power source, which runs on top of the ground plane. The graph below shows the variation in received power with antenna length, both with the antenna in the beaker and out of it.

Figure: Power Received with Antenna Length.

We measured the power transmitted in air by holding the antenna in the same orientation outside the beaker, at the same range from the antenna. In this first experiment, we damaged two transmitters. We did not turn off the transmitters when we cut back their antennas, so they may have been damaged by contact with our metal snips, as we describe above.

[23-OCT-08] We repeated the above experiment, this time with a epoxy-encapsulated transmitter with pins and a steel antenna wire like this. We turned off the transmitter while we cut short the antenna. We placed a 500-ml beaker of water 15 cm from our A3008C antenna. We stood the transmitter straight up in the water, resting upon the bottom. We used a straight antenna wire. We moved the beaker aside for the measurements in air. Peak power in water occurs at roughly 30 mm in both experiments. A 30-mm antenna transmits 20 dB more power out of our 500-ml beaker water than it does in air.

A 915-MHz quarter-wave antenna over a large ground plane will be 81 mm long in air and 9 mm long in water. Our antenna is more like an asymmetric dipole, with the battery forming one side and the wire forming the other.

Flexible Wires

Of three transmitters encapsulated with Process X that ION implanted in live animals in February 2008, all failed through wires snapping. In Process X, the wires are soldered to pins that emerge from an epoxy encapsulation. The point at which a wire enters the solder joint is weak because stress concentrates at the point where the effective diameter of the conductor drops from that of the joint to that of the wire. Two silver wires that broke near the animal's skull, but the remaining seven wires broke where the wire emerged from the solder. Among these seven were two silver wires, two tinned, stranded copper wires, and two solid copper wires. We saw no severe corrosion of the wires.

We conclude that finding the right wire for long life in an animal will be a project in itself, and we report upon that project in Flexible Wires. We conclude that teflon-insulated, 125-μm 316 stainless steel wires are our best choice for both the analog input leads and the antenna.

Live Animals

[20-FEB-08] Three transmitters implanted in rats in ION, London, provided robust transmission at range 50 cm. During 20 seconds of data from two moving rats simultaneously, we obtained 95% message reception 90% of the time, and 80% or greater reception the remaining 10% of the time. These transmitters failed after a day or two when their solid-conductor, copper and silver wires broke from repetetive stress.

[16-MAY-08] Pishan Chang of ION, London, implanted one transmitter with steel wires and epoxy encapsulation in a rat on 09-MAY-08. For a photograph of an identical transmitter, see here. On 15-MAY-08 she began taking hourly data from the transmitter, recording its average X value and number of messages received in a one-second period. From the average X we estimate the transmitter battery voltage. From the number of messages received, we deduce the percentage of messages transmitted that were received correctly. We plot battery voltage and reception percentage below.

Figure: VBAT and Reception from Implanted Transmitter.

[20-MAY-08] At hour 120, we find that 95% of our samples show reception of 80% or greater, which is sufficient to support our 160-Hz bandwidth. Pishan believes the poor reception 5% of the time is due to unfavorable orientation of the animal.

Figure: Rat in Cage with Implanted Transmitter. Loop antenna is in the background.

[21-MAY-08] We receive the picture above from Pishan at ION. We suspect the metal lid of the cage is responsible for some poor reception orientations. With the rat standing in the far corner from the antenna, the reflection off the lid will be strong and opposite to the line-of-sight signal. For a close-up of the animal, see here. For a view of the ION animal laboratory see here. The laboratory appears to be free of metal obstructions near the cage, and is therefore well-suited for robust reception.

[23-MAY-08] The battery voltage has begun a gradual decline, which we have already observed when we are roughly half-way through the battery's operating life. Our hope is that the battery will last another four weeks, or 700 hours. We now have almost 200 hourly data points, of which only 9 exhibit less than 80% message reception.

[27-MAY-08] Battery voltage has dropped to 2.1 V. The transmitter will fail soon. We are at first disappointed, but then we add up the total current consumption from this particular battery as follows. First, there was a 100-hr burn-in in January with the transmitter in salt-water. Then the transmitter ran for 100 hours at ION while its wires broke off. It ran for 100 hours before ION began taking data in this most recent experiment, and 300 hours during data taking. That's 600 hours of active life total, for 80 μA × 600 hr = 48 mA-hr. The battery was fresh on 27-DEC-07, so it has provided quiescent current for 3000 hours, which consumed 20 μA × 3000 hr = 60 mA-hr. That's a total of 108 mA-hr. The battery capacity is 120 mA-hr. Perhaps we have forgotten another 100 hrs of active life somewhere along the way.

[29-MAY-08] Transmitter is still running. Battery voltage is 2.05 V.

[02-JUN-08] Transmitter failed at 2.00 V after a total of 675 hours of operation and 3000 hours sitting on the shelf. The wires survived 475 hours in the animal without failure. The battery voltage appears to rise after it descends to 2.0 V, and then decend again. This apparant rise in battery voltage may be real, or it may be a failure in our measurement of battery voltage, which is based upon the average value of the analog input. The ADC is not guaranteed to work below 2.7 V. We find it works well down to 2.3 V, but at 2.0 V, it may fail to convert properly, giving rise to an apparant increase in battery voltage when in fact the battery voltage is dropping.

[AUG-08] The rat containing the transmitter has doubled in size. ION removes the transmitter. Its analog leads have snapped at the animal's neck. ION implanted another transmitter in a live animal. The transmitter ran for two weeks, recording seizures with 95% message reception. After two weeks, the rat escaped from its cage and squeezed into a cupboard through a narrow gap. The transmitter reception afterwards was poor, and the recorded voltage jumped from one end of the dynamic range to the other.

[15-SEP-08] Received two transmitters from ION. The first is the one that resided for twelve weeks in a rat that doubled in size. The analog wires snapped half-way along their length. All three wires were broken at the base, but still held in place by their coating of silicone dispersion. The analog leads broke at the solder joints. The antenna broke 1 mm from the solder joint. We suspect that these leads broke because the rat doubled in length. Even the antenna would then be under stress because its loop would be held in place by fibrous tissue.

The second transmitter we received is from the rat that escaped its cage. Both its analog wires and the antenna are snapped off at the base. It appears that water penetrated under the silicone and smudged the ink on the ID label. The battery is run down to 30 mV. We know that these leads broke while the rat was loose from its cage. Reception was perfect before, and failing afterwards. We suspect that the strain put upon the transmitter by passage through a narrow gap with the rat's body pulled the wires off their joints.

EEG

The first successful implant of a transmitter was in July 2007 by ION. The transmitter resided in a rat for three weeks, with X+ protruding down into its brain, and X− soldered to a grounding screw. In the third week, ION provoked epileptic seizures in the animal by injecting kainate into the rat's blood. Data recording took place in a room adjacent to the animal laboratory, owing to a power cut in the animal laboratory. According to Matthew, this room was full of boxes and instruments. We'll call it the cluttered room. Data reception in the cluttered room was unreliable. When the animal rolled over onto its back during a seizure, reception stopped alltogether. The figure below shows how the number of missing messages varied with time, along with the amplitude of X.

Figure: Missing Messages and Signal Amplitude versus Interval Number. Recorded in cluttered room, in which reception is unreliable. Interval length is one second. Most intervals are consecutive, but there are several gaps of a few minutes each. We expect 512 messages from the A3013A in each interval. When the number of missing messages exceeds 300 in one second, we suppress the amplitude of the signal to zero, and so remove spurious amplitudes from bad messages.

Although this data comes from the first successful implant, this was not the first implant. Previous implants in the animal laboratory had ended with the transmitter failing after several hours or days. While the implanted transmitters were working, however, signal reception was good, with no more than 50 messages being lost per second even during seizures.

Despite frequent loss of signal, there are many seconds during which we received all messages from the transmitter. The figure below shows an example of one of the quieter moments during the recording, and one of the loudest. The rms amplitude of the quiet signal is around 50 μV, and of the loud signal is around 500 μV.

Figure: X Quiet (Left) and Loud (Right). Vertical divisions are 200 μV, horizontal divisions 100 ms. For the raw data see quiet and loud.

Of more interest to neurologists are the following traces, which we provide together with their fourier transform, as produced by the Neuroarchiver. According to Matthew Walker, "The first demonstrates high frequency fast activity at the beginning of the seizure and the second demonstrates repetitive spikes at 2-3 Hz with superimposed faster activity."

Figure: Fast (Top) and Slow (Bottom) Oscillations. As seen in Nueroarchiver. Voltage versus time is on the left, amplitude versus Frequency is on the right. Note that the frequency scales are different. For the raw data see fast and slow.

The fast oscillations show a sharp peack in the frequency spectrum at 35 Hz, with a sharp second harmonic at 70 Hz. But there is no third harmonic at 105 Hz, nor a fourth harmonic at 140 Hz. If either of these harmonics were present in the signal, we would see them with the A3013A's X input, because its bandwidth is 160 Hz, as shown here.

After three weeks with the transmitter implanted, the rat died. For raw data recorded over five seconds from the recently-deceases brain, see here.

Conclusion

When encapsulated in epoxy, and equipped with stailness steel wires, the A3013 provides robust communication at ranges 50 cm and up in both London and Boston. The transmitter body is rugged. It is unaffected by vacuume and entirely water-proof. Its steel wires are immune to corrosion in an animal's body fluids, and are hundreds of times more resistant to fatigue than silver wires. When active and transmitting 512 samples per second, the A3013A will operate continually in an animal's body for at least 1500 hours, which is almost nine weeks. With a slightly larger battery, the same circuit will run for over four months.