The use of infrared (IR)
light as a means of wireless communication between computers, computer
peripherals, digital cameras, and other consumer products has gained wide
acceptance in recent years.
This is primarily due to the low cost of implementing IR solutions in contrast to radio-based implementations. The increasing pressure to produce low-power, high-speed consumer products in this arena, however, makes the implementation of IR transceivers, which is an integrated transmitter and receiver, more challenging. This article will address some of the key technical issues that need to be considered when designing IR transceivers.
Before discussing the design of the transceiver, it is important to understand the system requirements for IR wireless communication. In 1993, the Infrared Data Association (IrDA) was formed to create interoperable, IR communication protocol standards.1 Currently, the IrDA has standardized two modes of operation for IR communications: serial IR (SIR) and fast IR (FIR). The SIR mode supports data rates from 9,600 bps to 115 kbps. The FIR mode supports data rates of 576 kbps, 1.152 Mbps, and 4 Mbps. Outside of the IrDA organization, a few other notable IR communication protocols have been introduced. One of these protocols, introduced by Sharp, is called Sharp amplitude shift keying (ASK). Sharp ASK supports data rates from 2,400 bps up to 38.4 kbps. The other protocol, introduced by Apple, is called Apple-Talk; it supports data rates up to 230.4 kbps. Figure 1 shows the encoded modulation schemes for various IR communication protocols.
As shown in Figure 1, for data rates up to and including 1.152 Mbps within the IrDA protocols, the presence or absence of a single pulse of IR light within the bit period is used to signify a 0 or 1, respectively. The IR communication protocols illustrate that the transceiver must be able to receive and transmit optical pulses of varying duration and duty cycle. The transmitted optical power should be no more than the minimum required to ensure a reliable communication link, since power consumption in a battery-operated device needs to be minimized. Furthermore, the transmitted optical power must not exceed the allowable energy limits (AEL) for eye safety specified by the IEC-825.1 standard for LED and laser products.2 As a result, the receiver requires maximum sensitivity to achieve maximum link distance. Since the transmitter itself may lie anywhere from zero distance to the maximum link distance away from the receiver, the input dynamic range of the receiver has to be large. Due to the nature of the user environment, the receiver must be able to filter out interference from ambient light sources, such as sunlight, incandescent light bulbs, and fluorescent lamps, while maintaining the dual requirement for sensitivity and dynamic range.
Table 1 details some of the more important IrDA specifications, which are used as the framework to obtain design specifications for the transceiver. Figure 2 shows a system-level block diagram of an IR transceiver. It shows a structure for the receiver where the photocurrent from the detector is processed through a preamplifier, postamplifer, and comparator/slicer in a differential manner. A postamplification stage after the preamplifier may be necessary to ensure that sufficient signal amplitude is achieved for use by the comparator. The comparator is used as an A/I) converter to enable interfacing with a digital modem. It does this by comparing the signal to the voltage on the threshold input. This threshold voltage level is generated by the level detect block, which measures the strength of the receive signal and adjusts the threshold in order to maximize the bit-error rate (BER). Figure 2 also shows the gating of a light-emitting diode (LED) driver bythe TX line from a modem, and illustrates how an LED may be controlled to transmit an optical pulse.
LED and photodiodes
The basic optoelectronic components required are the LED at the transmitter and the photodetector at the receiver. IrDA specification require that the optical channel operate in the near-infrared spectrum from 850 nm to 900 nm. The sensitivity of the photodetector is a function of the wavelength of the incident light. To maximize efficiency, the LED should be chosen so that its peak emission wavelength lies near the peak sensitivity of the photodiode. Figure 3 graphs the spectral sensitivity of a typical silicon photodiode and indicates the peak wavelengths of various types of LEDs. The figure shows how GaAs LEDs (e.g., Temic TSHF5400) are nicely matched to a silicon photodiode (e.g., Temic BPV22NF). Beyond spectral matching, it is important when choosing; an LED to consider other properties such as its radiation angle, radiant intensity, and rise and fall times. In general, choosing an LED is relatively straightforward. The output intensity of the LED is directly related to the amount of current flowing; through it.
The choice of a photodetector however, requires more effort, as its performance is significantly affected by the interfacing circuitry.
For high-speed optical communications, photodiodes are preferred, given their superior frequency response. Photodiodes (PD) are usually operated under a receive-bias voltage, and can be modeled by the circuit shown in Figure 4. The main photocurrent iS is generated through the creation of electron-hole pairs when photons from the incident light penetrate the diode. There is a linear relationship between the photocurrent, iS, and the irradiance, Ee which is a measure of the intensity of the incident light and is given in W/m2. The photocurrent is calculated by:
1 ) is=Seff(l)Ee
where Seff is the effective sensitivity of the diode in units of A*m2/W. The sensitivity is a function of the wavelength, and takes into account the spectral sensitivity of the diode as well as the effect of the device's lens. As an example, the Temic BPV22NF is rated to provide 85 nA of photocurrent per 1 uW/cm2 irradiance. In practice, it appears that the sensitivity is, to some extent, a function of the frequency of the signal. Thus, when comparing photodiodes, the designer should test both the static (dc) and dynamic (ac) performance of the devices.
The current source, in , models the inherent noise of the photodiode, principally the shot noise generated by the DG leakage current and the photocurrent generated by ambient light. The noise is white in spectrum and has a spectral density of:
2) N(f) = 2qIs A2/Hz
where q = 1.69 x 10-19 C , and Is is the dc component of is. The root-mean-square (rms) value of in can be calculated by taking the square root of the product of the spectral density Is, and the bandwidth of the received signal.
The remaining two elements of the model, Rs and Cd represent the series resistance of the diode and the diode capacitance. The series resistance is a fixed value; for the Temic 8PV22NH; for instance, RS = 400.
Since the diode is operating under reverse bias conditions, the capacitance Cd is dominated by the depletion capacitance across the PN junction. As a result, Cd is greatly dependent on the applied reverse bias voltage for the BPV22NF photodiode. This characteristic is particularly significant when designing low-voltage receivers, as a low supply voltage severely limits the maximum reverse bias that can be applied to the diode. This ultimately impacts the frequency response of the receiver, and so the designer must ensure that sufficient steps are taken to achieve the required performance. For instance, one can potentially use a voltage multiplier circuit to bias the diode over and above the supply voltage. However, extreme care is needed to ensure that the added circuitry does not in itself introduce problems, such as injecting noise to the receiver.
TABLE 1: Specifications/requirements
for an IrDA compatible transceiver.
(9,600 bps to 115.2 kbps)
(576 kbps to 1.152 MBps)
|Min. optical receiver sensitivity (µW/cm2)||4.0||10.0||10.0|
|Max. optical signal on receiver (mW/cm2)||500||500||500|
|Min. receiver sunlight tolerance (µW/cm2)||490||490||490|
|Min. receiver fluorescent light
(µW/cm2 swept from 20 KHz to 200 KHz)
|Max. bit-error rate||<10-8||<10-8||<10-8|
|Transmitted optical power (mW/Steradian)||40-500||100-500||100-500|
|Peak wavelength from transmission/reception (nm)||850-900||850-900||850-900|
It is important to ensure that the applied reverse-bias voltage is constant. Poor regulation allows high frequency fluctuations to be capacitively coupled to the preamplifier input. Similarly, low frequency variations, due to temperature drift or supply voltage changes, alter the diode capacitance, which in turn can change the overall response of the system. One solution is to integrate a linear regulator as part of the diode's biasing circuitry. This will keep the constant, ideally independent of temperature or supply voltage.
Having discussed the system requirements along with the photodiode behavior in the previous sections, it is now possible to derive the electrical design specifications for the receiver, which are summarized in Table 2. From the table, one can observe a number of challenges in the design of an IR receiver. First, the extreme sensitivity requirement for the receiver demands a receiver noise that is low enough to handle photocurrents as small as 160 nA at the required BER. Secondly, the need to handle signal currents that vary from 160 nA to 42 mA represents a dynamic range of 108 dB, all in the presence of strong ambient lighting.
The dc photocurrents produced by ambient light can be filtered out by a simple high-pass filtering of the photocurrent. Fluctuating but periodic photocurrents, such as those produced by fluorescent lamps, may pass through due to the low-speed requirements of communication protocols such as IrDA's 9,600-bps rate. Fortunately, most of the optical power emitted by fluorescent lamps is in the visible spectrum and not in the infrared regions. A more serious problem is the level of shot noise introduced by the dc currents flowing through the photodiode. As the shot noise is white in its spectral density, much of this noise current is impinged upon the receiver. In the overall receiver chain, shot noise from the photodiode, noise from the detector's biasing circuits, and the noise from the first stage of the receiver amplifier become the dominant contributors of noise. The maximum sensitivity of the receiver, therefore, depends on minimizing the total amount of noise at the preamplifier. As a result, the preamplifier structure will be examined next before returning to the issue of maximizing receiver sensitivity.
Many circuit configurations exist for preamplifiers. The most popular and well-suited configuration is the transimpedance amplifier shown in Figure 5. The transimpedance amplifier converts an input current to an output voltage. For a large open-loop gain, Ao, the transimpedance of the amplifier, which is defined as the closed-loop, current-to-voltage gain, is given by:
where is the input signal current and Rf is the feedback resistor. Thus, the transimpedance of the amplifier is simply the feedback resistor value. The input resistance, Rin, of the resulting amplifier is given by:
The inherent noise of this amplifier topology arises from the feedback resistor, and from the input-referred noise currents and voltages of the amplifier itself. Resistor, Rf, is a source of thermal noise having a spectral density of:
The noise of the amplifier and the photodiode bias circuit is dependent on the device and circuit topology used. As such, the proper selection of devices and circuit configuration can enhance the noise performance. MOSFET and bipolar junction transistor (BJT) devices, for example, vary greatly in their noise characteristics. MOSFET devices exhibit a thermal noise current density of:
where K is Boltzmann's constant (1.38x10-23 J/K), T is the Kelvin temperature and gm is the transconductance of the FET in operation. MOSFETs also exhibit a low-frequency 1/f noise current of:
where Ids is the drain-source current, Cox is the oxide capacitance, and W*L is the gate area. In addition, thermal noise contributions exist for any resistor, poly-gate, or substrate resistances, and have a noise density that can he determined based on Equation 5. In BJT's, the primary source of noise arises from shot noise due to the base currents, where the shot noise current density, similar to Equation 2, is given by:
where Ib is the dc current flowing through the base. In addition, the BJT also has resistive noise components where the base resistance is the primary contributor to the noise.
TABLE 2: Receiver's system and electrical design specifications for IrDA compatibility.
|Peak signal Irradiance Ee||4µW/cm2||500mW/cm2|
|Sunlight irradiance Ee||490µW/cm2|
|PD sensitivity Seff (nA/µW/cm2)||40||85|
|Resulting input signal current||160 nA||42 mA|
|Photocurrent due to sunlight||42 µA|
The above equations address
the various sources of noise and their device dependence. The final decision
on the choice of the devices used is fairly complicated. In addition to
noise, other factors such as circuit configuration and frequency response
play a role in the overall use of the devices. BJT devices, for example,
provide better gain and have higher bandwidths, while MOSFET's have better
high-frequency noise charac-teristics. The core amplifier may hence use
a combination of both FET's and BJT's in a BiCMOS implementation. As a
result, obtaining the noise contribution from the amplifier itself is design
dependent . Any reader interested in further pursuing this discussion should
refer to the work of Steyaert, et al.4
To illustrate the analysis process, we will now describe the worst-case
sensitivity for a particular IR receiver. A typical noise current of 5
nA will he used for an amplifier with a 15-MHz bandwidth. The sensitivity
analysis is then performed using the simple noise model for the transimpedance
amplifier and its input noise currents.
In this example, root-mean-square (rms) values are assumed in all calculations. For a transconductance Rf= 2Ok over a 15-MHz bandwidth, Equation 5 gives a noise current of 7 nA. The realistic noise current of 5 nA is assumed for the core amplifier as stated before, and a similar estimate of 5 nA for the photodiode bias circuit's noise current is made. The shot noise from a photodiode with a dc photocurrent of 42 mA (due to 490 mW/cm2 of sunlight and a worst-case, maximum sensitivity of 85 nA/µW/cm2) would be 14 nA. Since each of these noise components is created independently it is reasonable to assume that they are uncorrelated. consequently the combined noise current can be obtained by taking the square root of the sum of the squared currents:
To calculate the sensitivity of the receiver, a number of assumptions need to be made about how the modem determines whether a 0 or 1 bit is received. If a peak detection scheme; is used, for instance, the modem may simply sample the output of the comparator during the midpoint of each bit period; the sampled output, then, depending on the modulation protocol, would be decoded as either a 0 or 1. If we assume that the noise is Gaussian with zero mean, then the optimum BER is achieved by setting the threshold of the comparator at half the peak amplitude, ipeak of the IR pulse. This maximizes the Euclidean distance between the threshold and the ideal symbols representing 0 and 1. An example implementation of a level detect circuit that sets this threshold level can be found in the work of Nakamura, et al.5 The BER for this type of system is well understood and is given by:
where erfc ( ) is the complementary error function from statistical theory, and the rms value, intotal is the noise's standard deviation. Solving the previous equation for a minimum BER of 10-8 as required by the IrDA gives us the minimum signal current required:
This would translate to a minimum detectable signal of 4.8 µW/cm2 of incident optical power if a worst-case photodiode sensitivity of 40 nA/µW/cm2 is assumed. However, recall from Table 1 that the minimum receiver requirement is 4 µW/cm2. Hence, the above amplifier configuration would fail to meet the sensitivity specification! This example shows the challenge in designing a receiver with the required sensitivity.
The next issue that the designer needs to address is the 108-dB dynamic range of the input signal. The sensitivity requirements for the receiver demand that the preamplifier have a large feedback resistor, Rf, since a large value for Rf maximizes the signal gain while lowering the noise contributed by the resistor. Unfortunately, the larger the transimpedance, the lower the dynamic range, because it becomes easier for the amplifier to be saturated by strong signals. A larger transimpedance also demands a proportionally larger open-loop amplifier gain, Ao, in order to maintain the same bandwidth. This can be seen from Figure .5 where the dominant frequency pole, p, at the receiver input is given by:
In order to keep p constant, therefore, Ao/Rf must also remain constant. Thus, the designer is faced with conflicting requirements for sensitivity and dynamic range.
A number of circuit techniques can be used to deal with improving the input dynamic range of the receiver, thereby alleviating some of the restrictions imposed on the transimpedance of the amplifier. Figure 6 illustrates three possible techniques: variable transimpedance, diode clamping in the feedback loop, and input current shunting. The variable transimpedance amplifier operates by reducing its closed-loop gain for large input signals. This is done by obtaining the average value of the voltage output and then letting this control the value of the transimpedance in the feedback.7 In practice, the variable resistor can he implemented using NMOS transistors operating in the triode region. The use of a clamping diode in parallel with the maximum transimpedance value for the amplifier essentially means that the diode branch, with its lower series resistance, Rf, becomes the dominant value once the output exceeds the clamping voltage of the diode branch. This in turn ensures that the output of the amplifier dues not saturate. Finally, the use of a current shunting circuit before the amplifier allows variable attenuation of the input signal based im the signal level. Figure 6 illustrates how the virtual ground at the negative opamp input can be used to attenuate the signal current, is by a factor of R1/(R1 + R2). This ensures that the amplifier does not have to deal with excessively large signal currents and, hence, the transimpedance value of the amplifier can be kept constant and at its maximum. The amount of current shunted is once again controlled by the signal level. The ability to control the values of R1 and R2 based on the signal level allows for a wide attenuation factor that can thereby increase the input dynamic range of the receiver. Receiver dynamic ranges of 73 dB have been reported using this technique.8
In all the schemes above, the gain of the receiver can become non-linear for intermediate and large signal values, especially during the transition or sampling time of the first few output pulses. When a clamping diode is used, the pulse shape is always distorted for signals that are strong enough to activate the diode branch. The pulse shape distinction is caused by the high gain of the amplifier during the low-valued portions of the input current and the low gain of the amplifier during the high-valued portions of the input current within the same data pulse. Consequently, the slicer or comparator circuit that follows the receiver will produce pulse widths that exceed the actual signal pulse widths. Since sampling of the first few pulses is required for the current shunting and variable transimpedance schemes, the pulse-width criteria demanded by the interfacing digital modem cannot be restrictive during this sampling period. Fortunately, many of the high-speed data protocols that demand strict pulse-width adherence also have preamble sequences that enable the receiver to linearize its gain if the variable transimpedance or current shunting scheme is used.
The transmitter design, in principle, is fairly simple since it only requires the gated switching of a constant current through the LED for optical transmission. An elementary design may consist of a large common-source NFET device whose drain is connected to the cathode of the LED while the gate is connected to the transmit signal. The gating transmit signal in turn is provided by the interfacing modem. The anode of the LED is then connected either directly or through a current-limiting resistor to the power supply voltage. The optical power, Po, which is radiated, is measured in mW/Steradian. It is linearly related to the amount of current, ILED, flowing through the LED as stated by:
where K is the optical efficiency of the LED. Typical values for K are 400 mW/ASr, as in the case of Temic's TSHFS4OO LED. The design of the transmitter circuit is fairly simple if thermal power dissipation is not of concern. However, in battery -operated devices, the power supply voltage that feeds the LED can vary by as much as 2V as the battery discharges. This requires some form of current regulation through the LED to ensure that a minimum amount of optical power is produced under all conditions and a maximum amount is not exceeded in compliance with the IEC-825 standards on eye safety. As stated earlier, thermal power dissipation becomes the greatest challenge in this implementation, since the NFET driver size has to be maximized to reduce power dissipation in the driver, and proper heat-sinking on the LED is required if it is integrated as well. For example, in an application that requires a Po of 200 mW/Sr, 500 mA of peak current has to be pumped through the LED. For a drain-to-source voltage (Vds ) of 200 mV across the driver FET and a forward diode voltage drop of 1.8 V across the LED, the peak power dissipated is:
For a protocol that has a duty cycle of 1/4 (as in IrDA's FIR), this translates to an average power of 250 mW. In a typical plastic package with a thermal resistivity, Rth, of 160 C/W, this would raise the silicon die temperature over the ambient by:
The challenge in the transmitter design is ensuring that the Vds of the FET is minimized over the range of rail voltages, and that the package is designed for optimal heat dissipation.
The existence of various low-speed and high-speed IR communication protocols, the need for extreme photo-sensitivity (< 4 µw/cm2), the requirement to accommodate a 108-dB dynamic input signal range, the presence of ambient IR light sources, and the need for the receiver to operate over a range of voltages from 2.5V to .5V make the design of a receiver in an IR transceiver very challenging. The transmitter circuit in a transceiver, in turn, needs to ensure that a constant optical power is generated over the range of operating voltages, and that thermal dissipation is optimized. This article has addressed the key transceiver design issues that need to be considered to resolve the specified criteria. In the process, we have outlined techniques that may be used in an actual implementation.
Future trends for higher data rates and the continual demand for lower operating voltages, such as 1.5V will make IR transceiver design for high-speed communication challenging for years to come.
Ravi Ananth is a staff engineer in the wireless analog ASIC group at the IBM Microelectronics Lab in Toronto, Canada. Ananth received his M.A.Sc. in E.E. from the university of Toronto.
Mark Noll is the development manager for the wireless ASICs group at the IBM Microelectronics Lab in Toronto Canada. He obtained his B.Sc in E.E. from the University of Florida.
Khoman Phang is currently pursuing his Ph. D. degree at the University of Toronto. His interests include high-speed analog integrated circuits, adaptive filtering, and wireless communications.
This article appears with
permission from Communication Systems Design, October 1997 Copyright 1997
Miller Freeman Inc. All Rights Reserved