Figure: Encapsulated Subcutaneous Transmitter with Steel Wires.
In our Technical Proposal, we layed out our plan to develop a transmitter small enough to be implanted in the body of a rat, fast enough to transmit four hundred data samples per second, powerful enough to be detected at a range of three meters, and efficient enough to operate for three months on a lithium battery. We divided our development into four stages.
Stages One, Two, and Three we completed with success. The electronics of Stage Four were ready in March of 2007. It was then that we confronted the problem of robust, water-proof encapsulation for the transmitters. At the time we wrote our Technical Proposal, we believed that two coates of silicone dispersion would be adequate to protect the transmitter from water. Given the simplicity of the procedure, there would be no point in coating the transmitters ourselves. We proposed to send transmitters to our users without batteries, wires, or coating. Our user could solder their preferred wires to the transmitters, install the battery, and apply the coating. When the battery ran down, they could strip off the silicone, replace the battery, and coat the transmitter again.
But water-proof encapsulation is not straightforward. Capillary action makes water a relentless invador of any opening in the encapsulation. Air bubbles trapped beneath the transmitter components emerge into the curing silicone and cause microscopic cavities. The large air bubble beneath the battery expands when the external pressure drops, forcing out the coating. In the low pressure of an aircraft cargo hold, such bubbles burst right through six coats of silicone.
Our work on encapsulation is an appendix to the work we committed to in our Technical Proposal. After nine months of effort, we arrived at an encapsulation process that uses both epoxy and silicone, which we call Process X. We made our first transmitters with Process X in December 2007. We describe our work on encapsulation and encapsulation methods in our Encapsulation report.
Once our transmitters were water-proof, they endured for long enough in live animals for the wires to break from repetetive stress. The wires are bend repeatedly by the animal as it moves around and turns its head. Wires broke at the neck of a rat, and at their solder joints. When we combine this repetetive stress with the requirement that the wires operated well as an RF antenna, and resist corrosion in body fluid, we arrive at the problem of choosing flexible wires for a long-life implanted device.
Our work on flexible wires is an appendix to the work we committed to in our Technical Proposal. After three months of effort, we selected stainless steel, teflon-insulated wires. We made our first transmitters with steel wires in May 2008. We describe our work on wire fatigue in our Flexible Wires report.
In the sections below, we describe the performance and functioning of our Subcutaneous Transmitter system. If you can't understand our terminology consult our Terminology page, where we define the terms RF, IF, LO, mixwer, attenuator, dBm, and dB. We explain the significance of 50 Ω in our circuits, and we provide a table that relates power in 50-Ω circuits like our to signal amplitude.
Here we list the most prominent improvements we made in Stage Four when compared to Stage Three.
One thing we tried in the course of our Stage Four development was slowing down the bit rate in the transmitter messages. We believed that we could increase the operating range by decreasing the modulation bandwidth and increasing the signal's spectral power density. We present a comparison of the system's performance at the two bit rates below. We found that the slower bit rate increased our operating range by roughly 30%, but decreased the battery life by 30% and doubled the frequency of transmitter collisions in multiple-transmitter systems. We did not think the loss of battery life and increased collisions were worth the 30% increase in gain, so we decided to stick with the same bit rate we used in Stage Three. An added benefit of keeping the same bit rate is that our Stage Four transmitters and receivers are compatible with those of Stage Three.
Our Subcutaneous Transmitter (A3013) provides two analog inputs, one or both of which can be enabled. One input is intended for high gain and high bandwidth. The second is intended for low gain and low bandwidth. We imagine the former monitoring neural activity, and the other monitoring heart rate.
Our Data Receiver (A3018) comes in a 9" × 7 " × 3" metal box. The box connects to our data acquisition system (LWDAQ) through an RJ-45 connector. It receives power through the same connector. A BNC socket provides connection for a coaxial cable to the antenna. There are four lights on the box. One of these, a green one, turns on when the Data Receiver receives messages from a transmitter.
Our Antenna (A3015) attaches to the end of a coaxial cable through a BNC socket. We recommend a 96" cable (240 cm). The antenna stands up on its own, and can take one of three forms: loop, dipole, or whip.
The uncoated Subcutaneous Transmitter (A3013) is 16 mm × 18 mm × 7 mm.
Our receiver is larger than necessary. We designed it for ease of use, construction, experimentation, and modification. Inside the box are three circuits: the Demodulating Receiver (A3017), SAW Oscillator (A3016SO), and Data Recorder (A3007C). The receiver's antenna input, which is a BNC socket in the side of the box, connects to the RF input of the Demodulating Receiver. The output of the SAW Oscillator connects to the LO input of the Demodulating Receiver. The output of the Demodulating Receiver connects to the input of the Data Recorder. The receiver connects to the data acquisition system through a single RJ-45 socket. Power for all three of the internal circuits arrives through this socket.
| A3013 | Subcutaneous Transmitter |
| A3014MT | Modulating Transmitter |
| A3014SO | SAW Oscillator (Precise) |
| A3015A | Loop Antenna |
| A3015B | Dipole Antenna |
| A3015C | Whip Antenna |
| A3016SO | SAW Oscillator (Stable) |
| A3017 | Demodulating Receiver |
| A3007 | Data Recorder |
| A3018 | Data Receiver (Collection of Circuits in an Enclosure) |
We can disconnect our Antenna from the Demodulating Receiver and instead connect some other RF source with known power and frequency. We can vary the power we deliver with the help of fixed attenuators in series with the cables. We made use of our Modulating Transmitter (A3014MT) as a sweeping source of RF to measure our RF amplifier frequency response. We used our 910 MHz SAW Oscillator (A3014SO) as a stable source of RF power that we varied from +10 dBm down to -80 dBm. We used the same 910 MHz osxillator as a source of RF power for a transmitting antenna, and so tested radio transmission at a known frequency. We switched antennas quickly, from loop to dipole to whip, using our BNC connectors. We used our 868±1 MHz SAW Oscillator (A3014SO) as a frequency reference to calibrate all our other oscillators. Any time we doubted a result, we could switch all the circuits with duplicates, to see if the result was sound.
We spent some time developing and checking several ways of measuring RF power, as we explain elsewhere in a section called Power Measurement. The procedure we settled upon in the end was to downshift the RF signal with our ZAD-11 mixer, and measure the IF power. We obtain the original RF power by adding the ZAD-11's insertion loss, and our cable losses, to the IF power. This gives us a measurement accurate to ±1 dB.
When we measure RF signal frequency, we compare frequencies to our reference 868±1 MHz SAW Oscillator (A3014SO) with the help of a mixer. All our frequency measurements are calibrated with respect to this 868 MHz reference. We mix the RF signal whose frequency we want to measure with the 868 MHz, and measure the IF frequency using our oscilloscope. The precision of our IF frequency measurement is 0.5 MHz or better.
And so it is that we can measure the power of RF signals to ±1 dB (10%) and the frequency of RF signals to ±1 MHz (0.1%). In our previous work, our power measurements were no better than ±3 dB, and our frequency measurements were all referred to the center of our SAW filter passbands, which turned out to be off by up to 5 MHz from their nominal centers. Our increased accuracy, combined with the convenience of our BNC cables, fixed attenuators, and mixer, allow us to measure the performance of our RF circuits in more detail than before.
The Data Recorder (A3018) plugs into a Long-Wire Data Acquisition (LWDAQ) system. It acts as a LWDAQ device, although it violates the current-consumption limits of the LWDAQ.
We use the Data Recorder with the LWDAQ Driver with Ethernet Interface (A2037E). The A2037E connects to the Internet, your Local Area Network, or directly to your computer via an RJ-45 Ethernet socket. You communicate with the A2037E, and therefore the Data Recorder, via TCP/IP.
On the computer you use for data acquisition, you run the LWDAQ software, which you can download from here.
Download the latest version of the LWDAQ software from here. To help you with installation and use of the LWDAQ software, consult the User Manual. With the Subcutaneous Transmitters and Data Receiver, you will use the Recorder Instrument and Neuroarchiver Tool. Please note that the Recorder and Neuroarchiver in LWDAQ 6.5+ are compatible with the Data Receiver (A3018), but incompatible with the Data Receiver (A3010).
The Recorder Instrument is a set of routines that run inside the LWDAQ program. You can use the routines by opening the Recorder Instrument window from the Instrument menu. You can call the routines from the LWDAQ's console. The Recorder Instrument software uploads blocks of binary data from the Data Recorder hardware and divides them into blocks of fixed time-duration. It displays transmitter signals as it receives them, each transmitter trace in a different color, and prints a summary of the received signals to the screen.
The Neuroarchiver Tool uses the Recorder Instrument to upload signals from the Data Receiver. The Neuroarchiver uploads, filters, displays, and stores to disk selected signals from the Data Recoder hardware, and does so using the Recorder Instrument as an intermediary. The Neuroarchiver calculate and display the fourier transform of the incoming signals. It stores data and transforms to disk.
The Subcutaneous Transmitter (A3013) provides two analog input channels by means of four flexible, insulated copper wires. One or both channels will be active, depending upon how the device has been programmed. Each channel is high-pass filtered with an RC network that rolls off at around 0.1 Hz. The primary input channel is low-pass filtered with a three-pole active filter with roll-of anywhere from 10 Hz to 320 Hz. The auxilliary channel is low-pass filtered with an RC network at anytwhere from 1 Hz to 320 Hz.
Each version of the A3013 provides a different combination of sampling rate, dynamic range, and channel activation. The A3013A, for example, has two wires for one active channel with dynamic range 9 mV, bandwidth 160 Hz, and sampling rate 512 SPS. The A3013A can detect a 5-μV sinusoid of any frequency between 5 Hz and 160 Hz, and measure its amplitude with 1-μV accuracy.
The A3013 converts its amplified and filtered inputs into sixteen-bit numbers and transmits them in radio-frequency messages.
The A3013 transmits messages through its RF antenna. The A3013A transmits in the 902-928 MHz ISM band. It transmits at two frequencies to represent two logic levels. The messages propagate through space as radio waves. (See How Antennas Work in our Technical Proposal for an explanation of antennas and radio waves.)
The transmitter's analog circuits operate continuously to amplify and filter its analog signals. But the logic and RF circuits are inactive almost all the time. In the case of our A3013A, the logic and RF circuits wake up for 7 μs every 1953 μs, which is 0.36% of the time. During this 7-μs burst of activity, the transmitter converts its single analog input into a sixteen-bit number, and transmits this number, together with some synchronizing bits, the transmitter's four-bit ID, and a checksum. We call this transmission a message. The A3013A's message rate is 512 per second. Within each message, the transmitter sends bits at the bit rate. The A3013 bit rate is 5 MBPS, or 200 ns per bit, just like its predecessor, the A3009, and its transmissions are fully-compatible with the Stage Three electronics. For more details of the message encoding, see below.
The radio messages are received by our Antenna (A3015). The messages are joined in the antenna by RF signals from other sources of similar frequency. The messages and the interference propagate along a coaxial cable to the Data Recorder (A3018).
The radio signal enters the metal enclosure of the Data Recorder and connects to the RF input of the Demodulating Receiver (A3017) inside (see schematic). The RF signal, messages and interference, are amplified by 20 dB. They pass into a bandpass filter that rejects signals outside the 902−928 MHz ISM band. What remains are the messages and ISM-band interference. These are amplified by another 20 dB and enter a mixer. The mixer downshifts the 902-928 MHz ISM-band RF signal to a 38-64 MHz IF signal.
The IF signal passes through three limiting amplifiers. Each provide 22 dB of gain, and limit their output. The output of the third stage is called IFL in the schematic. When we disconnect the antenna, the signal on IFL is random, but not quite saturated. When we connect the antenna, the RF interference in our laboratory causes IFL to saturate almost all the time. Because IFL is saturating, its amplitude is constant. Only its frequency changes.
The fixed-amplitude IFL passes into a tuned circuit called the discriminator. The discriminator attenuates the lower IF frequencies, so that its output amplitude depends upon the frequency of IFL.
The discriminator output passes into a full-wave demodulator. The demodulator turns the AC signal into a signal proportional to the AC amplitude. The demodulator output is D in the schematic. The alternating radio frequencies of the transmitter messages appear in D as a square wave.
The D signal passes along a short BNC cable to another circuit inside the Data Receiver box, the Data Recorder (A3007C). There it is mildly band-pass filtered by two RC networks to form the signal S in the schematic. The transmitter messages appear in S as a square wave centered about the 0-V potential.
A comparator transforms S into a logic HI when S is above 0 V, and logic LO when it is below 0 V. The comparator output is called C. The logic levels of the original transmitter message now re-appear as logic levels on C. Outside the messages, C is a random or pseudo-random sequence of logic levels. It is random when the dominant source of RF power at the antenna input is electronic noise, and pseudo-random when the dominant source is interference in the ISM band.
The C logic signal passes into the Data Recorders's logic chip, where it is sychronized with the message clock. Our message clock is 40 MHz. The synchronized version of C is called SC in the logic chip's firmware. The SC signal changes only on the rising edges of the message clock, so changes pulses on C that take place between these edges do not appear in SC.
The Data Recorder constantly monitors the stream of bits on SC, looking for messages embedded within the stream. These messages could arrive at any time. Messages can also appear in the random SC bit sequence by chance. These random messages we call bad messages, and we discuss them in detail below.
The Data Recorder stores any messages it detects in its 512 KByte first-in first-out buffer as thirty-two bit records, as we describe below. The last eight bits are a timestamp, which counts cycles of the Data Recorder's 32.768 kHz clock.
The Data Recorder also stores clock messages 128 times per second. The clock messages takes the form of a transmitter message from a transmitter with ID zero. We reserve ID zero for clock messages, and will never be use it as the ID of an actual transmitter. The sixteen-bit data of a clock message is a sixteen-bit counter that increments by one with every clock message stored. The final eight bits are a timestamp, which is always zero, because the clock messages are stored whenever the Data Recorder's 32.768 kHz cycle counter wraps around to zero. You can think of the clock messages as being the product of a virtual transmitter with ID zero, transmitting 128 messages per second, and whose data is a counter that increments from one message to the next.
The Data Recorder connects to the LWDAQ. The LWDAQ in turn connects to an Ethernet. Your data acquisition computer runs the LWDAQ software and communicates with the LWDAQ by TCPIP.
The LWDAQ software on your data acquisition computer uses its Recorder Instrument to upload blocks of messages from the Data Recorder's first-in first-out buffer. It plots these messages on the screen, including the clock messages. The Recorder provides you with blocks of data that cover a time interval you specify. As it acquires data, the recorder makes no effort to eliminate bad messages or insert substitute messages. Its data blocks are therefore of variable memory-size, but fixed time-duration.
The Neuroarchiver Tool calls the Recorder to supply fixed time-duration data blocks. You tell the Neuroarchiver which channels you want to display and record. Each channel corresponds to a transmitter identifier. Our Stage Four transmitters can have identifiers between 1 and 15 (0 is reserved for clock messages). The Neuroarchiver reconstructs each channel from the data block by removing any bad messages it can identify, and inserting substitute messages where necessary. It displays the reconstructed channels, and calculates their fourier transforms.
The Neuroarchiver will store the raw Recorder data blocks to disk, or it can store reconstructed channels to disk in a variety of formats. It will store fourier transforms to disk also. And so the transmitter signals are available on disk for you to analyze as you see fit.
The bit rate is the rate at which the transmitter sends bits during one of its message transmissions. The A3013A, for example, transmits a total of 40 bit-values in 8 μs. The bit rate is 5 MBPS (megabits per second). We describe the function of these bits in the Message Encoding section of an earlier report.
With the exception of a few temporary, slower versions we made for the following tests, all Stage Four and Stage Three circuits run at 5 MBPS. Here we compare the performance of the Stage Four system at 5 MBPS and 2.5 MBPS and show that the gain in operating range we make by dropping the bit rate is slight, while the loss in operating life is significant.
We set up a Demodulating Receiver (A3017) with a SAW Oscillaotr (A3016SO) and Data Recorder (A3007C). We connected a Dipole Antenna (A3015B) on a 240-cm (96") coaxial cable (RG58C/U). We recorded messages from a Subcutaneous Transmitter (A3013). We configure the transmitter and Data Recorder for message transmission at 5 MBPS (200 ns per bit), then for 2.5 MBPS (400 ns per bit). The switch between bit rates requires the following changes.
We measure a variety of performance parameters for each bit rate. We present our measurements in the following table.
| 2.5 MBPS | Parameter | 5 MBPS |
|---|---|---|
| 918±4 MHz | Frequency Modulation | 918±4 MHz |
| 13 MHz | Bandwitdth (90% Power) | 18 MHz |
| 12 μs | Message Duration | 8 μs |
| 108 μA | Active Current | 75 μA |
| 18 μA | Inactive Current | 18 μA |
| 15 m | Maximum Range (favorable orientation) | 10 m |
| <0.1/s | Bad Message Rate (Antenna Disconnected) | <0.1/s |
| <0.1/s | Bad Message Rate (Transmitter Inactive) | <0.1/s |
| <0.2% | Blocked Message Rate (Transmitter Adjacent to Antenna) | <0.2% |
| ≈3% | Missing Message Rate (Transmitter Moving Randomly 1 m from Antenna) | ≈2% |
| ≈15% | Missing Message Rate (Transmitter Moving Randomly 1 m from Antenna and Enclosed Between Two Hands) | ≈25% |
The 5 MBPS and 2.5 MBPS transmissions perform best with a modulation depth of ±4 MHz. The 90% power bandwidth of the 5 MBPS messages is 18 MHz, which leaves us with at least 8 MHz of extra room in the 902-928 MHz band. The power bandwidth of the 2.5 MBPS messages is 13 MHz. In theory, the narrower the bandwidth, the less likely a hole in the reception caused by interference (such as the this) will lie within the transmission bandwidth, and therefore the less likely that the transmission will be corrupted by interference.
The biggest benefit of the lower bit rate is a 50% increase in operating range.
The biggest costs of decreasing the bit rate are the increase in active current. The current rises from 75 μA to 108 μA. The operating life drops from nine weeks to six weeks.
Another cost of decreasing the bit rate is the increased transmission time, which means a greater probability of collisions between multiple transmitters sharing the same receiver. The probability of collision between any two transmitters on any given message transmission is 0.7% at 5 MBPS and 1.4 % at 2.5 MBPS. With ten transmitters, each would lose 7% of its samples at 5 MBPS and 14% at 2.5 MBPS. Our system can handle a 30% loss of messages. At 2.5 MBPS we would use up most of this missing message budget in the normal operation of ten transmitters.
We decided to stay with a 5 MBPS bit rate. Our collaborators at ION stated that battery life and bandwidth were more important than operating range, because in either case the operating range appears to be greater than the size of a standard rat cage.
Our Stage Four circuits use the same message encoding as the Stage Three transmitters, which we describe in the Message Encoding section of an earlier report.
We considered using a cyclic redundancy check at the end of the message, instead of the inverse of the transmitter ID for our error-checking bits. The cyclic redundancy check would allow us to detect corruption of data bits. The inverse-id bits are sensitive only to corruption of the ID or the bits themselves. But we decided the benefit of improved error checking did not outweigh the disadantage of incompatible transmitters.
The Data Recorder (A3018) stores these messages as four-byte records. The first byte is the transmitter ID. The next two bytes are the sixteen-bit data. The last byte is a timestamp. The timestamp counts periods of the Data Recorder's 32.768 kHz clock oscillator.
The Data Recorder can fail to identify the message in its incoming bit stream (see Signal Path), even though they are received clearly by the Demodulating Receiver. The Data Recorder might be occupied with something that looks like a message, and when the real message arrives, all it knows is that the previous message was invalid. The recorder goes back to looking for a new message, but is too late to identify the real message. We measure the rate at which the Data Recorder loses messages by putting a single transmitter next to the antenna and watching for missed messages. Our signal is strong and there are no other transmitters to collide with it. We assume that any missing messages have been lost in the Data Recorder. It turns out that the number of synchronizing bits that the Data Recorder requires at the beginning of a message is what dictates the missing messages rate.
| Number of Synchronizing Bits | Missing Message Rate |
|---|---|
| 2 | 1.2% |
| 4 | 0.5% |
| 5 | <0.2% |
Only after it has received its required number of synchronizing bits does the Data Recorder proceed to message recording, and it will lose an incoming message only if it is already recording a false message. By insisting upon five synchronizing bits, we ensure that the Data Recorder loses less than one in a five hundred messages. As it turned out, we settled upon 11 synchronizing bits because this improved the performance of the system when the signal was weak.
Each transmitter, when active, transmits messages in a continuous stream. The A3013A transmits 512 sixteen-bit data samples per second. A message that the Data Recorder fails to receive is a missing message.
The Neuroarchiver Tool handles missing messages by creating a substitute message with equal data sample value to the previous message received from the same transmitter. It does this only for the channels you are recording.
Despite the Neuroarchiver's message substitution, missing messages degrade the effective bandwidth and quality of our received signals. But they do not cause catastrophic problems, nor do they confuse our data acquisition system, which can rely upon a guaranteed stream of messages from each transmitter it monitors.
The primary cause of missing messages is lack of signal strength at the antenna. The primary cause of loss of signal strength is cancellation of the RF signal by its own reflections arriving at the antenna. We discuss the effect of reflections in our Operating Range section.
Another cause of missing messages is corruption of one message by a message from another transmitter, which will occur when two transmitters happen to choose the same time to transmit. We discuss collisions between transmissions in our Collisions section.
Noise and interference can generate messages on their own. We call these bad messages. Bad signals can raise false alarms, spoil the scale of self-adjusting displays, and wreak havoc with your fourier spectrums. We have made every effort to avoid them.
Bad messages have a valid transmitter ID number, but their data is invalid. The Recorder Instrument rejects them when they occur with ID numbers you have not asked it to record.
The bad message rate in our laboratory is less than one per ten minutes when there are no transmitters active. In London, the bad message rate was roughly one per second, sometimes coming in bursts of ten in a second. With transmitters active, we find that some messages get corrupted, as we describe below, and these message become bad messages.
We spend the rest of this section showing how bad message can, in theory, arise from noise and interference, and how our message encoding makes such bad messages very unlikely.
As we describe in Signal Path, the Data Recorder monitors a logic signal called SC, looking for messages. These messages could arrive at any time. If the stream of bits is random, there is a chance it will generate a message by chance. If the stream of bits has some pattern to it because of radio interference, the chance of a message appearing in it at random might be much higher.
When we disconnect the Data Recorder's antenna, and replace it with a 50-Ω terminator, we stop radio signals from reaching our receiver. All that remains at the input to our RF amplifier is white noise, whose power in our radio-frequency passband is around −87 dBm. The receiver amplifies this noise until it is large enough to generate an active pattern of zeros and ones on SC. This alternation is random, because the white noise is random. The output of the Demodulating Receiver (A3017) is an analog signal, S. This signal is sharply bandwidth-limited by the Demodulating Receiver's radio-frequency passband, and mildly high-pass filtered when it enters th Data Recorder. The Data Recorder uses a comparator to generate the logic level SC from S. When we examine SC on an oscilloscope with the antenna disconnected, the bandwidth of the bit stream is around 10 MHz, which we determine with the help of various low-pass filters.
Our messages represent individual bits values as edges in SC, not as levels of SC. An edge is a transition from logic LO to HI or from HI to LO. A rising edge is a one, and a falling edge is a zero. A sequence of synchronizing one-bits at the beginning of the message tells the Data Recorder when to look for data-carrying edges, because it knows the rising edges in the synchronizing sequence are the data edges in these initial one-bits. Every subsequent data-carrying edge allows the Data Recorder to adjust its expectations for when the next data-carrying edge will arrive. With a 5 MBPS bit rate, the data-carrying edges are separated by 200 ns. Each bit takes roughly 8 periods of our 40-MHz message clock. We say "roughly" because the bit rate is not exact. The transmitter's message clock is generated by a ring oscillator, and is accurate to only ±10%.
If the Data Recorder detects an edge within one or two clock periods of another, it rejects the entire message and goes back to looking for synchronizing bits. It rejects synchronising bits using the same contraint. The Data Recorder tests the level of SC for premature edges twice during each bit period. In order for a random sequence of bits to create a valid message, it must satisfy both tests for all bits of the message. There are around 25 bits in the message, so the random stream must take on the correct value 50 times. Furthermore, the message must have the correct checksum at the end, which is the same as saying that it must take the correct value for another five bits.
The likelyhood of a random sequence on SC matching our message protocol is roughly 0.530. To the first approximation, there are 105 opportunities for the match to occur every second, because each message occupies roughly 10 μs. We expect our bad message rate from noise and random interference to be of order on per hour. And indeed this is what we observe: after running for ten minutes we saw no bad messages with the antenna disconnected.
A corrupted message is one that has been interfered with, but which still passes through our error-checking. A corrupted message has the same effect upon data acquisition as would a bad message with the same ID.
We can reduce the likelyhood of message corruption by checking for errors in the content of the message, as opposed to only by comparing the final four bits to the first four bits. A four-bit cyclic redundancy check at the end of all transmitted messages would give us better rejection of corrupted messages. Instead of ending up as bad messages, most corrupted messages would end up as missing messages. But the cyclic redundancy check would make our Stage Four transmitters incompatible with those of Stage Three, so we decided to tolerate a higher than necessary corrupted message rate for the sake of backward-compatibility.
When two transmitters send a message at the same time, we say that they collide. In almost all collisions, the result is the loss of both messages. If the first message is much more powerful than the second, the second will be lost, and therefore act as a missing messages. The first message will be received correctly. But if the first message is weaker than the second, the second will interfere successfully with the first, while it is unlikely that the second will be received properly itself. When the two signals are equal in power, they will certainly interfere with one another.
Collisions occur because two transmitters are transmitting at the same time. If the moments of transmission were exactly regular, collisions could occur systematically at every transmission instant. But they are not regular. Each transmitter displaces its moment of transmission by a small, random, amount of time. This displacement is a fraction of the transmission period, but still several times greater than the time consumed by the act of transmission itself.
We call the displacement of the transmission instant transmission scatter. We describe how the data acquisition system handles transmission scatter in our Recorder Manual. We show how the hardware implements the scatter in the Transmission section of our A3013 Manual.
We define operating range as the greatest range at which reception succeeds in 95% of all orienations. Our definition of reception success is when we receive 80% or more of transmitted messages. An A3013A transmits 512 messages per second, so the operating range is the greatest range at which we receive 410 or more messages per second in 95% of a large set of randomly-chosen orientations and positions at the operating range.
When we hold a transmitter between two fingers and move it around randomly at range 1 m for a minute (see movie), we receive 98% of the messages it transmits. We estimate that the operating range of our transmitters, when held in air between two fingers, is 2 m in our laboratory. In Matthew Walker's office in London, we found the operating range to be closer to 1 m. Earlier interference spectra indicate that interference in Matthew Walker's office is stronger than in his animal laboratory, so we expect the operating range in the laboratory to be more like 2 m. Operating range in our OSI office in the center of Waltham is closer to 70 cm.
When we implant a transmitter in an animal, we expect to lose transmission power. The antenna will not be resonant, and power will be absorbed by animal tissue. Our earlier observations suggest that this loss is no more than 3 dB. If we lose 6 dB (three quarters of the power) our operating range will halve. We expect to receive at leat 95% of messages transmitted from a live rat in a cage.
We compared the operating range of the Data Receiver (A3010B), which is the Stage Three receiver upgraded with new firmware and an antenna socket, and the Data Receiver (A3018A), which is the original Stage Four receiver. (The original Stage Three receiver, the A3010A, suffered from a firmware bug that reduced its performance even farther with weak signals.) We used the same Subcutaneous Transmitter (A3013A) for both receivers. We used the same Loop Antenna (A3015A) for both receivers. We unplugged it from one receiver and plugging it into the other.
It appears that our Stage Four development has doubled the effective range of the transmitters. We claim that we are nearing the physical limits of message detection, as we shall now explain.
The noise in our receiver is the thermal and amplifier noise at the antenna input. The interference is RF power in the receiver's pass-band arriving at the antenna from transmitters other than those the receiver is intended to receive. Measurements of noise and interference in our demodulating receiver reveal that the effective noise power at our RF input is −90 dBm. But interference power, even in our basement laboratory, is over a hundred times more powerful, at −68 dBm.
We previously determined that our RF signals must have power 12 dB greater than the interference in order to avoid corruption. When we transmit power across 1 m of space from a quarter-wave antenna to a loop antenna, we observe that we lose at least 32 dB compared to connecting the power directly to the receiver circuit with a cable.
Even with our omni-directional antennas an unfavorable relative orientation of the transmitting and receiving antennas causes a 17 dB drop in received power. We must also deal with reflections, or multi-path interference, which we have discussed in all our previous reports, in particular in the Multi-Path Interference section of our Stage Three report, and the Radiated Power section of our Stage Two report. Reflections of the main signal interfere at the receiver. We can easily get a 10-dB loss due to destructive interference. In other words: we can lose 90% of our power easilyOur transmitters produce roughly −4 dBm. At a range of one meter in air, we will receive no more than −36 dBm. If the two antennas are unfavorably oriented, we will lose another 17 dB, so our received power drops to −53 dBm. This −53 dBm is still more than 12 dB above our interference power of −68 dBm. With these considerations alone, we expect reception to remain reliable in all orientations at 1 m.
But we must contend with multi-path interference as well, which give rise to reception dead spots. With the antenna in an unfavorable orientation, in which we receive only −53 dB from line-of-sight transmission, we have power radiating more effectively in other directions. This power reflects off nearby conducting surfaces and arrives at the receiving antenna. It arrives with a more favorable orientation and direction then the line-of-sight signal. This reflection adds to the line-of-sight signal. It might reinforce the line-of-sight signal, or cancel it, depending upon their relative phase.
The phase difference between a reflected and line-of-sight wave is a strong function of frequency. If the reflected path length is three meters long, this is a hundred wavelengths. A 1% change in frequency will cause a 2π change in phase. A 0.25% change in frequency will cause a π/2 change in phase. The figure below shows two holes in the response of our our Demodulating Receiver (A3017). The hole is about 1 MHz wide. The antennas are 1 m apart, on either side of our body.
We see that each hole is about 1 MHz wide. A 0.1% change in frequency causes the cancellation to stop. We take such sharp holes in the frequency response to be evidence of multi-path interference. The width of the holes is consistent with a reflected path length of several meters. Longer path lengths would give holes that were even more sharp.
Aside: The two holes are separated by 24 MHz, and you will notice yet another hole, on the left side, which is 24 MHz below the middle hole. As the frequency changes by 24 MHz, the phase difference between the line-of-sight and refelected waves changes by 2π. The number of wavelengths in the path difference changes by one. Because the frequency changes by 2.5% to bring about this one-wavelength change, the path difference must be 40 wavelenghts, or 12 m. This suggest to us that the wave is bouncing off one of the metal shelves five or six meters from our transmitting antenna. If that's the case, then the reflected signal will be about 25 dB weaker than the line-of-sight signal, except for the fact that our body is in the line of sight. Human tissue attenuates 900 MHz by approximately 1 dB/cm, so we expect a loss of around 25 dB through a human torso. Both signals are each strong enough for reception, but they cancel one another at particular frequencies.
With 10 dB loss due to cancelation by reflections occurring at the same time as 17 dB loss due to poor antenna orientation, our signal drops to −63 dBm, which is only 5 dB above our −68 dBm interference. Our signal will be dominated by interference.
If, on the other hand, we can find a favorable orientation of the transmitter, the line-of-sight signal will be stronger than any reflection, and we can expect −36 dBm at 1 m, and − 56 dBm at 10 m. Even at 10 m, a favorable orientation will give us a signal 12 dB above our interference, so we can expect reception. In fact, we find we can get reliable reception at up to 15 m in our lab.
Our experiments with the Subcutaneous Transmitter (A3009) showed us that we needed to increase the gain of its analog input from ×30 to ×300 if we were to see normal EKG activity in a mouse brain. We also detected a couple of sources of noise in our analog-to-digital conversion process. Apart from the increase in amplification, and the addition of an auxilliary input, the Subcutaneous Transmitter (A3013) is much the same as its predecessor.
The A3013, with its greater gain, has roughly 10 μV of cross-talk and thermal noise on its input. It can detect a 5-μV 160-Hz sinusoid with no difficulty, once you take the fourier transform of its output in the Neuroarchiver.
We studied the sources of noise in the analog inputs, and came to a clear understanding of mains hum, and how to reduce it.
For our work on encapsulation of the transmitter, see Encapsulation. We arrived at the encapsulation shown here. Casting epoxy cured with the help of a vacuum chamber provides a compact and water-proof encapsulation that resists a vacuum. Penetration of the encapsulation for electrical contacts is via gold-plated copper pins. Because these pins are stiff and encapsulated by epoxy, water cannot penetrate through them or around them.
For our work on flexible wires for the antenna and analog connections of the transmitter, see Flexible_Wires. We concluded that teflon-insulated, stainless-steel wires were the best compromise between ease of implantation and resistance to fatigue and corrosion. Our first prototypes use 125-μm diameter, full-hardened, 316-stainless wires. Later prototypes may try 75-μm wires of the same material. For a photograph of the wires, see here. We solder the wires to their pins with the help of acid flux. We coat the solder joints with three layers of silicone dispersion. The dispertion is water-resistant, and ensures that the electrical path from one pin to another through the animal body fluid is greater than 10 MΩ.
Battery life of the A3013A-E is roughly nine weeks running continuously in a live animal. For details of how the battery life is affected by battery size and transmission rate, see here.
[21-MAY-08] A single A3013A-E has been running continuously in a live animal since 05-MAY-08 at ION in London. They started recording hourly data from the transmitter on 15-MAY-08.
The rat with the implanted transmitter is shown above on 21-MAY-08. You can see the loop antenna in the background. It has diameter 95 mm. The cage is roughly 20 cm wide, 40 cm long, and 15 cm high. The top of the cage is made of metal. In theory, reflections off the metal grating when the rat is at the far end of the cage can prevent proper reception. With the rat standing on its hind legs in the far corner of the cage, for example, the reflected waves and the line-of-sight rays will be 180° out of phase with approximately the same path length. Nevertheless, performance with the metal lid is robust: we get 80% or more of the messages 95% of the time. For graphs and more photographs, see Live Animal in the transmitter manual.
The operating range of the A3013A in live animals is greater than 50 cm in Boston and London. The A3013A-E is encapsulated in epoxy and equipped with steel wires. The encapsulation is water-proof, vacuum-proof, and corrosion resistant. The wires are flexible, resistant to fatigue, and resistant to corrosion. Operating life in a live animal is limited by the capacity of the battery. The A3013A will operate continuously for nine weeks, transmitting 512 16-bit samples per second. The dynamic range of the A3013A input is 10 mV, with noise of around 10μV in its 160-Hz bandwidth. In quantities of ten or more, OSI prices the A3013A-E at around $400.
Operating range in both locations is limited by the geometric properties of our antennas, the strength of local interference, and the local arrangement of reflecting surfaces. There are only two ways we can increase the operating range of the transmitters. One is to increase the transmitter power, so that we can better dominate interference. If we increase transmitter power, we add components and reduce battery life. The other is decreasing transmission bandwidth, so we are less likely to be struck by multi-path interference. If we decrease our bandwidth, we increase transmission time, which in turn increases battery life. We are satisfied with the compromise our A3013 transmitter makes between between operating range and operating life.
Reception of the transmitter signals is with a simple loop antenna connected by a coaxial cable to our A3018A Data Receiver. The same loop antenna will receive signals from up to ten transmitters with only minimal loss of transmitter data through collisions. With the loop antenna at the center of four cages, the A3018 will receive and record data from the ten rats sharing the four cages. In single quantities, OSI prices the A3018A at $2000.
The A3018A Data Receiver communicates with a LWDAQ Driver with Ethernet Interface, A2037E. In single quantities, OSI prices the A2037E at $1000. The software to communicate with the A2037E and the A3018A is open source, and free to all user under the GNU Public License.
We conclude that OSI can provide a telemetric system for ten rats in an animal laboratory for only $7000. The transmitters will run for nine weeks. At the end of their life, they must be discarded. Their water-proof encapsulation cannot be recycled. A new set of ten transmitters costs another $4000.
Given the superior performance of our telemetric system, we are confident that these prices will be competetive.
