We are working upon the design of a battery-powered transmitter small enough to reside in the body of a rat. Our chief concern in the design is the power consumption of the transmitter circuit. The specified lifetime of the transmitter is three months of continuous operation, transmitting 400 SPS (samples per second) out of the rat's body to a receiver up to five meters away. The largest battery we can include in the 2-ml maximum specified volume of the tranmsitter is a 3-V Lithium battery with capacity 120 mA-hr.
The theoretical lower limit to the power consumption of the transmitter would be the minimum power it must launch into space in order for the receiver to pickup a signal strong enough to be identified over thermal noise. If we assume 400 SPS and 40 bits per sample (includes a 16-bit data value and some identificatio nbits), then our data bandwidth is roughly 10 kHz. Thermal noise power at the base of an ideal receiver antenna is 4.k.T.B, where k is Boltzmann's constant, T is absolute temperature, and B is bandwidth. In our case, the thermal noise would be about 1 fW (one femptowatt, or 10^-15 W). If the transmitter launches 1 nW into the space around the rat, an antenna with effective aperture 100 cm^2 at range 5 m will pick up 30 fW, which would, in theory, be enough to swamp the noise. In theory, our transmitter could consume only 1 nW. With perfectly efficient electronic components, the transmitter could run off a 120-mA-hr battery for fourty thousand years.
Sadly, we do not have perfectly efficient electronic components. Only by using low-power components released in the last twelve months do we have any hope of achieving a three-month lifetime by frequency-modulated radio transmission of digital samples. The transmitter's average current consumption must be less than 60 uA.
Another constraint upon us when we choose to transmit information with radio waves is that the wavelength of our carrier wave must be small enough to allow us to place an efficient transmitting antenna in the rat's body. We have chosen to work at roughly 1 GHz, for which a quarter-wave antenna length above a ground plane in free space is roughly 80 mm long. Because the dielectric constant of salt water, and therefore a rat's body, is roughly 80, a quarter-wave antenna in a rat's body, with effective salt-water all around it, might be as short as 80 mm divided by the square root of 80, or 9 mm.
If we use a 1-GHz carrier, then we might also transmit our data bits at tens of megabits per second (MBPS), in which case we could transmit our data in bursts. If we can find a low-power 1-GHz VCO (voltage-controlled oscillator) with sufficiently high modulation bandwitdth (that's how fast you can change the output frequency of the oscillator), which we can shut off and turn on again whenever we need to transmit a sample, then we might be able to keep the current consumption of our RF (radio frequency) transmitter down to a few tens of microamps.
Maxim, a semiconductor company, has a line of low-power VCOs. We picked the MAX2624. When transmitting, it consumes 8 mA, and produces 500 uW of 1 GHz power into a 50-Ohm load (that's 160 mV rms). You can vary the frequency between roughly 950 MHz and 1000 MHz by driving the TUNE input from 0.4V to 2V. You can put the chip to sleep by driving the !SHDN input to 0V. So, that looks perfect, all we need to do is connect MAX2624's output to a quarter-wave antenna (a length of wire sticking out from our circuit) and we should be able to launch at least half the power into space. We can then trasnmit our data bits by changing the frequency of the oscillator (frequency modulation), which we do by driving the MAX2624's TUNE input with a logic gate. Our receiver detects the changes in frequency and so receives our data bits. It would be nice if we could turn on the transmitter, transmit our bits at a rate of 100 MBPS, and then turn off the transmitter, all in less than a microsecond. That way the average transmission current consumption would be 8 mA for 1 us at 400 SPS, or 3 uA. But we would be happy to complete the transmission in 5 us.
Unfortunately, Maxim does not tell us how long it takes the MAX2624 to stabilize after we bring it out of shut-down, nor how long it takes to turn off when we shut it down. Nor does Maxim tell us how fast the VCO frequency responds to changes in the TUNE input. They told us we had to do these tests ourselves. Because we could find no other VCO with a shut-down line and such low operating power consumption as the MAX2624, it seemed to us that the future of our rat transmitter design upon this particular chip's modulation bandwidth and start-up delay. The Modulating Transmitter (A3001A) section of our Rat Transmitter Test Board (A3001) allowed us to test both properties of the chip.
The Modulating Transmitter consists of a MAX2624, some resistors and diodes to deliver TUNE and !SHDN inputs, and a directional coupler and an antenna. The MAX2624 output enters the OUT pin (yes, the OUT pin, that's what they called it) of the directional coupler and emerges on the IN pin with hardly any loss of strength. After that, the 1-GHz wave proceeds to the base of the antenna. So far, the characteristic impedance of the path it has followed has been 50 Ohms. If we have a perfectly-tuned antenna sticking up from our circuit board, it will appear to the incoming wave as a resistor to ground (and when we say ground, we mean the zero-volt power plane of the circuit board). The value of this resistance turned out to be something like 75 Ohms when I came to tuning the antenna, perhaps because our ground plane is not so large, and appears to form the opposite side of a dipole antenna with our vertical wire antenna. Because the resistance seen at the antenna base is greater than fifty ohms, we can add a resistor to ground to bring the total resistance down to exactly 50 Ohms, so that the incoming wave flows forward without reflection. Indeed, we deduce the value of the antenna resistance by assuming that in parallel with the shunt resistance it will create a resistance of 50 ohms. Some of the power goes through the shunt resistor (R9 in the schematic) to ground, and the rest goes up into the antenna and propagates through space.
But before we have a perfectly tuned antenna with the correct shunt resistor, we will have an impedance at the base of the antenna that does not match the 50-Ohm characteristic of the 1-GHz wave. Consequently, some of the power of the wave will reflect from the antenna base and return to the directional coupler IN pin. Reflection is a problem for two reasons. First, it cuts down the power that the VCO generates, by messing with the VCO's output amplifier, and second, any power reflected from the antenna base is power that is not being transmitted into space. We need to adjust our antenna length and our shunt resistance to minimize the reflection, but this raises the question of how we are going to measure the size of the reflection, when both it and the outgoing wave reside on the same transmission line.
What we need is something that separates the reflected wave from the outgoing wave, and thus allows us to measure the relative size of the reflection. A directional coupler does exactly that. The directional coupler we use in our Modulating Transmitter is the ADC-20-12 from Minicircuits. A fraction of the reflected wave emerges on the coupled port (pin 3), accompanied by a far smaller fraction of the outgoing wave. Our circuit board carries the coupled signal to a BNC plug. If we have some way to measure the 1-GHz power emerging from the coupled port, we will be able to adjust our antenna length and shunt resistance until we minimize the reflected power.
The Modulating Transmitter also provides BNC plugs to bring in the !SHDN and TUNE inputs from our function generator. We can use these to test the modulation depth, modulation bandwidth, and turn-on delay of the MAX2624.
Our first problem, when trying to test the Modulating Transmitter was to figure out how we would be able to measure the relative amplitudes of 1-GHz signals, and also their relative frequencies. In our lab, we have a 300-MHz analog oscilloscope. We at first assumed that we could not make any measurements of 1-GHz signals with a 300-MHz scope, but we were wrong. We connected the RF output from the Modulating Transmitter to a BNC plug, and so launched the VCO's 150-mV rms output into a 50-Ohm coaxial cable. We ran this cable to our scope input, and set the scope to terminate the input with its own internal 50-Ohm resistor. We saw a 7-mV rms cloud on the scope screen. With the time base switched to 5 ns per division, we counted just under five oscillations per division. We were looking at about 900 MHz.
If we assume that our analog scope responds as if it had a low-pass filter with corner frequency 300 MHz, then the fact that our roughly 150-mV rms input shows up as a 7-mV trace at 900 MHz suggets that the scope response is dropping by roughly a factor of twenty for a tripling of input frequency. In going from 900 MHz to 1000 MHz with constant input power, we expect the amplitude of the trace to drop from 7 mV to 4 mV. In other words, our scope can measure the relative amplitudes of two signals of the same frequency, and the relative frequency of two signals of the same amplitude, despite the fact that the scope is specified only up to 300 MHz.
Our first experiment was to connect a triangle wave running from 0 V to 3 V to the T input of the Modulating Transmitter. We left the S input open so that !SHDN would float high, and enable the VCO. We connected the RF output to a BNC cable and ran it to our oscilloscope. Figure 1 shows both the triangle wave and the RF output. You can see that the frequency of the output does indeed change as the TUNE input varies, and the change in frequency is consistent with 0 V giving 900 MHz out and 3 V giving 1 GHz out. The MAX2624's modulation depth is therefore close to 100 MHz.
Figure 1: Output of the MAX2624 (top trace, 5 mV per division) as seen by our 300-MHz oscilloscope with a 0V to 3V triangle wave (bottom trace, 1 V per division) on the TUNE input. The timebase is 2 us per division. The change in output amplitude indicates a change in the output frequency.
Now that we have confirmed the modulation depth of the MAX2624, and our ability to measure relative changes in amplitude, we can connect our scope to the R output of our Modulating Transmitter, and so measure changes in the power reflected from our antenna. First, we leave the antenna disconnected and remove the shunt resistor. We know that under these circumstances, the antenna base reflects all incoming power back to the source, and so we determine the amplitude of the scope trace cloud that corresponds to 100% power reflection.
We connected a 20-cm piece of wire as an antenna, and we left the Modulating Transmitter set up as it was in the previous test, with a triangle-wave TUNE input and the RF output connected to the oscilloscope. We did not take any pictures of the oscilloscope screen as we tuned the antenna, but we did see large changes in the reflected power as we cut the antenna back from 15 cm to 5 cm. Because we were sweeping the VCO frequency from 900 MHz to 1 GHz, we were able to see, on the oscilloscope screen, the change in reflected power with frequency, and so minimize the reflected power as best we could across the entire frequency range. In the end, we found that the reflected power was minimized with a 10-cm vertical antenna, just a piece of insulated wire 100 mm long sticking straight up off the circuit board. When we put a 135-Ohm resistance to ground from the antenna base, the reflections decreased to less than 10% of the outgoing power (or -10 dB, where dB is for decibel, a logarithmic measure of power).
If less than 10% of the power incident at the antenna base is being reflected, and the shunt resistor is 135 Ohms, and the characteristic impedance of the wave travelling through the directional coupler is 50 Ohms, then conservation of energy appears to imply that at least 60% of the 1 GHz power produced by the MAX2624 is entering the antenna. There is nowhere for it to go after that but into space. If we assume that the MAX2624 does indeed generate 500 uW (-3 dBm) into 50 Ohms, then we conclude that we are radiating at least 300 uW into space. There is nowhere else for the energy to go. This 300 uW is two orders of magnitude more than we need to receive a 10 MHz signal at a range of 5 m with a crude dipole antenna.
We apply a square wave to the TUNE input and see how fast the frequency of the VCO output changes. This amounts to a measurement of the VCO's modulation bandwidth. Figure 2 shows a 10-MHz square wave applied to the TUNE input, and the consequent change in VCO output frequency. As before, the output frequency is indicated by the amplitude of the cloud in the upper trace. Our square wave input goes from 0 V to 3 V, which is the size of the square wave we expect to apply to the VCO in our battery-powered miniature transmitter. As you can see, the frequency of the output follows the TUNE input, settling to its new value in approximately 50 ns. This suggests that we can, without any complications, transmit 20 MBPS (mega bits per second) by frequency modulation.
Figure 2: Output of the MAX2624 (top trace, 5 mV per division) as seen by our 300-MHz oscilloscope with a square-wave (bottom trace, 1 V per division) on the TUNE input. The timebase is 50 ns per division.
We have established that our VCO, the MAX2624, can transmit 20 MBPS by frequency modulation. This means we can transmit an 80-bit message, containing 16 data bits, twenty-four identifier bits, and fourty receiver-synchronizing bits, in 4 us (four microseconds). Now we need to determine if we can wake up the MAX2624 from its shut-down state, in which it consumes less than 1 uA, and put it into stable oscillation, in which state it consumes 8 mA, in a matter of a few microseconds or less. We connected the T input of the Modulating Transmitter to 0V, thus setting the output frequency to about 900 MHz, and applied a square wave to the S input, thus driving the MAX2624 in and out of shut-down.
Figure 3 shows the square wave input and the VCO output. The amplitude of the cloudy part of the upper trace indicates a combination of the amplitude of the VCO output and its frequency. The output is stable within within 500 ns of !SHDN going to 3 V, and disappears in less than 100 ns. We conclude that the MAX2624 is ready to transmit within 500 ns, and shuts down within 100 ns.
Figure 3: Output of the MAX2624 (top trace, 5 mV per division) as seen by our 300-MHz oscilloscope with a square-wave (bottom trace, 1 V per division) on the !SHDN input. The timebase is 100 ns per division.
The MAX2624 provides us with 10 MHz modulation bandwidth, and turns on and off in less than a microsecond. We can wake up the chip, transmit eighty bits, and shut it down again in less than 5 us. If we do this four hundred times a second to support our 400 SPS sample rate, our average transmission current consumption is 16 uA (5 us out of ever 2.5 ms multiplied by 8 mA), leaving roughly 40 uA available for the rest of the transmitter circuit. With this remaining 40 uA we must filter the incoming neuron potential, amplify it, digitize it to sixteen-bit precision, and construct an eighty-bit sequence that conveys both the data and the transmitter's unique identity number, as well as synchronizing bits required by the receiver. We have found chips whose data sheets guarantee that they can do all of these things for us for a maximum of 60 uA, with a typical consumption of around 30 uA.
Provided that the MAX2624 does indeed produce 500 uW at 1 GHz, and we have no reason to suppose that the Maxim data sheet should be in error in this respect, we are confident that our Modulating Transmitter, which uses the MAX2624 as a source of 1-GHz RF power, is radiating 300 uW into space. With a crude dipole antenna, we will present an antenna aperture of at least 100 cm^2 to the radiating wave front, and at a range of 5 m we will receive at least 10 nW. If our receiver has a signal bandwidth (as opposed to reception frequency, which is in this case 1 GHz) of 10 MHz, the thermal noise power our signal must overcome will be 0.2 pW. So our expected signal to noise ratio at 5 m is 60,000, or 50 dB. The 10 nW received signal will develop 710 uV across a 50-Ohm antenna base impedance (-50 dBm).
Work done at Surrey University by Jeffries et al. suggests that we will lose 1 dB/cm at 1 GHz through a mammalian body. If a rat is 10 cm thick, we will lose about 10 dB, leaving us with 1 nW received power at a range of 5 m. We will also suffer a surface reflection loss of about 4 dB. Our signal to noise ratio will still be 36 dB.
Our chief concern in receiving the transmitter signal is not the strength of the signal itself, but its strength relative to other sources of 900 - 1000 MHz radio waves. These frequencies are used by cordless phones and mobile wireless instruments. We may find that transmissions from such devices corrupt the rat transmitter burst transmissions. If so, we may have to dictate that no such instruments be used near the laboratory. On the other hand, we may be able to narrow down the frequency band used by our transmitters, and so decrease the chance of interference.
We are now confident in our transmitter design, but we would like to demonstrate data reception over a five-meter distance using the MAX2624 as a transmitter before we accept an official purchase order from our London-based collaborators. We plan to perform these tests over the weekend and early next week. Chief among the tests is measuring the effect of cordless phones upon the receiver, and measuring the background 1 GHz noise in our laboratory.