Home Patents Binary Sampler Articles Contact
Stabilize the 74AC04 Limiter with a Tracking Threshold
Saturday, Dec 20, 2008
Table of Contents
- Introduction
- 74AC04 Limiter AM/PM Conversion
- Measuring The 74AC04 Transfer Function
- Cascading 74AC04's For More Gain
- Measuring The Open Loop Gain
- Closed Loop Step Response
- Summary
Introduction
The 74AC logic series gathers praise from people such as Ulrich Rohde for its excellent phase noise performance. One of the most useful devices is the 74AC04 hex inverter used as a limiter to convert sine waves to square waves for logic processing. As an analog part, it has many desirable features such as high input impedance, fast risetime, low propagation delay, low noise, low current drain, rail-to-rail inputs and outputs, single supply operation, and multiple sources. It is ESD protected, latchup protected, well documented, cheap and readily available, and will probably be around in one form or another well into the future.
What's not to like? One problem is the totem-pole output produces very fierce ground bounce and switching transients on the Vcc. A second problem is the switching threshold can vary a huge amount, from 30% to 70% of Vcc, which means it can be anywhere from 1.5V to 3.5V with a Vcc of 5V. This limits the maximum input signal to 3V p-p or +13.5dBm, and leads to a very subtle flaw discovered in some amazing engineering work by Martein Bakker, PA3AKE.
As Kevin Wheatley, M0KHZ, points out in his blog (updated here), Martein discovered a serious noise problem with the Analog Devices AD9910 1GHz DDS chip . Due to a flaw in the design, broadband noise from an internal bandgap voltage reference appears on the output signal as amplitude modulation.
Since the 74AC04 has such poor control over the switching threshold, the AM noise can be converted into PM noise, which degrades the otherwise excellent performance of the AD9910 chip.
One possible solution is to use a bifilar wound center-tapped ferrite toroid transformer to couple the differential sine wave signal to the input of the 74AC04, then stabilize the zero crossings with a simple feedback circuit as described in Fig 1 below. This will force the duty cycle at the output to match the voltage at the junction of R4 and R2, which is normally set to 50% of Vcc. A small trim capability might be desirable to account for harmonic distortion in the signal.
Fig 1. 74AC04 Limiter Stabilized with Tracking Threshold
74AC04 Limiter AM/PM Conversion
The design process starts by showing how the 74AC04 threshold can produce AM to PM conversion. The analysis uses Mike Englehart's excellent LTspice program available at http://www.linear.com/designtools/software/ltspice.jsp.
For the analysis, we model the 74AC04 using the tanh function in a behavioral voltage source (BV). This is shown below in Fig 2. The equation is of the form
V = 2.5*(1 - tanh(500 * (V(Va) - 2.5)))
The tanh function provides a differentiable output from -1 to +1. The inverter gain is set to 500, and it multiplies the input signal Va minus the threshold of 2.5V. The result is multiplied by 2.5, so when the input signal is zero, the output is 2.5V, and it varies from 0 to +5V as the input signal crosses the threshold.
Two circuits are shown to illustrate the results with a threshold of 2.5V and 1.5V.
Fig 2. 74AC04 Switching Thresholds
The resulting output signals are shown in Fig 3 below. The green trace is the signal at the input pin when the threshold is at 2.5V. The black square wave shows no asymmetry since the threshold is at Vcc/2.
Fig 3. Threshold Switching
The lower graph shows the input signal in red when the threshold is at 1.5V. Due to the feedback in a normal limiter circuit, the duty cycle of the output must change to produce an average output voltage of 1.5V. The resulting asymmetry is shown in the blue trace.
Since the output no longer switches at the zero crossings, it is clear that amplitude modulation on the input signal will phase modulate the output signal. A slight change in the amplitude of the signal will move the output transition sideways, which converts the AM noise into PM noise.
Measuring The 74AC04 Transfer Function
To add a tracking threshold, we need to analyze the 74AC04 as a pulse width modulator, or PWM. In order to generate a stable closed loop, we need to know the transfer gain as the input signal amplitude changes. The circuit to measure this is shown in Fig 4. The input sine wave is superimposed on a 5V ramp, and the output is low-pass filtered and compared to the ramp signal.
Fig 4. 74AC04 Transfer Function at 3V and 0.3V p-p
Fig 5 shows how the transfer gain changes with input amplitude. The blue curve shows the gain is close to unity for input signals of 3V p-p or more, and the green curve shows the gain increases as the signal amplitude decreases. You can show this is true logically since the output must switch from one rail to the other over a smaller range of input signals.
In an ordinary circuit, the increase in gain would limit the usable dynamic range, since the gain and phase margins would decrease as the signal amplitude decreases and the loop would go unstable at some point. However, this circuit has a robust open loop response, and the simulation was quite happy with input signals as low as 2mV p-p. This raises some messy issues about cascading stages to get more gain, and I doubt the actual hardware would accept amplitudes this low as gracefully. A 74HC04 might have a better chance, but it would take very careful attention to grounding and bypassing. The summary at the end has a link with more information on prototyping sensitive circuits.
Due to lack of time, I did not measure the gain at 0.3V p-p (green curve) since it is unlikely that amplitudes this low would be used in a low phase noise application.
Fig 5. Transfer Function at 3V and 0.3V p-p
Fig 6 shows the gain is approximately 2 at 2V and 4 at 1V p-p input.
Fig 6. Transfer Function at 2V and 1V p-p input.
Cascading 74AC04's For More Gain
One reason for using the behavioral voltage source (BV) instead of the LTspice 74AC04 logic device is so we can change the gain and see what effect it has on the transfer function. It turns out that once the gain is above a certain limit, further increases have negligible effect on the transfer function.
This means it is probably not necessary to cascade more sections to increase the gain. It won't have much effect on the transfer function except to increase the variation in propagation delay with changes in temperature, which we don't want in some applications.
However, if your application runs at lower frequencies where the 74AC04 output risetimes are too slow, it might be necessary to cascade more sections to get faster risetimes. Eventually you will reach a point where the 74AC04 starts oscillating due to the slow input signal risetime and internal impedance of the device. This may be solved by adding some form of hysteresis, or even substituting a 74AC14. However, hysteresis will make the limiter vulnerable to AM/PM conversion again, which won't be easy to overcome.
Fortunately, very low phase noise is seldom needed on low frequency signals, and this circuit can still be used to maintain a 50% duty cycle on the output signal when hysteresis or a 74AC14 is used.
Measuring The Open Loop Gain
Now that we know the 74AC04 transfer gain as a pulse width modulator, we can apply feedback around the circuit and measure the open loop gain. Fig 7 shows the circuit with the component values selected for optimum step response.
Fig 7. Circuit to Measure Open Loop Gain
The resulting open loop gain and phase is shown in Fig 8. This shows the closed loop response will be stable with the component values shown.
Fig. 8. Open Loop Gain and Phase.
Closed Loop Step Response
Fig 9 shows the circuit to measure the closed loop step response.
Fig 9. Closed Loop Step Response Circuit.
The resulting waveform is shown in Fig 10. The black trace is the 3V p-p signal at the input pin of the 74AC04. It is in series with a 1V square wave to perturb the circuit.
The blue trace is the output of the RC low-pass filter that averages the 74AC04 square wave output. This is applied to the input of the op amp, which produces an output voltage that will restore the 74AC04 to the desired 50% duty cycle. This is the signal shown in red, which is applied to the input transformer secondary, then to the input pin of the 74AC04. Divider resistors R2 and R4 were used to trim the damping. As shown, the output settles in about 4 milliseconds.
Fig 10. Closed Loop Step Response
Summary
The method described above provides a reasonable wide-band limiter with essentially zero drift in the switching threshold regardless of temperature, changes in Vcc, or device variation in the 74AC04. It is very difficult to find any other method that provides this stability.
Since the output signal is switched from Vcc to ground, any noise on the Vcc supply will appear in the output. You might be interested in various methods of filtering ripple and noise on supply voltages described here
If you are interested in taming the fierce switching transients of the 74AC series, here is a prototyping method that can help.
For those who wish to examine the analysis further, the LTspice files are archived here . The archive contains a dirlist.ndx file which lists all the files along with a brief description. The .ASC files contain the schematic and parameters. The .PLT files tell LTspice what signals to show on the graph. You don't have to know anything about SPICE to view the files. Just load the desired .ASC file into LTspice and press Alt-1. It will do the analysis and plot the results as shown in the above figures.
One of the most interesting demos is to load ACLIM03.ASC, ACLIM04.ASC, and ACLIM05.ASC into LTspice simultaneously. Click on each one in turn and press Alt-1 to run the analysis.
This will give you the open loop gain, the linear step response, and the PWM step response while tracking a 3V p-p input signal. The red trace in the ACLIM04 graph should agree fairly well with the blue trace in the graph for ACLIM05. Make sure they all have the same component values and gain. You can change a value by right-clicking on a component and entering the desired value.
It is most illuminating to change a component in all three files, and view the resulting change in open loop gain and phase margin, and watch how that affects the step response.