This basic low power modulator was described in the LT Journal Nov 2015 article. In this blog post, we will show how to:
- Use simulation to validate the filter design, and
- Reduce DC offsets by utilizing Differential Amplifiers for the baseband drive.
Also shown are some miscellaneous tips to help complete the design.
To generate complex modulation, two identical channels are required (I and Q). For simplicity, only one channel is shown here.
In this example design, the system specifications are as follows:
- Symbol rate= 30 ksps
- Digital filter alpha= 0.35
Next step is to determine the baseband filter requirement:
- Total signal BW = 30*1.35= 40.5 kHz, which is 20.25 kHz for each I and Q ch. This determines the filter passband.
- The filter stopband requirement is largely determined by the DAC update rate, in this case 240 ksps, or 8x interpolation per symbol.
Now let’s fast-forward to the complete circuit, followed by the description. Note that for simplicity, only one of two identical I & Q channels is shown:
Figure 1. Baseband drive circuitry converts Single-ended DAC to Differential. Passive DAC reconstruct filter is recommended for low noise.
DAC Input & Reference
Differential DAC outputs are preferred. But when single-supply, single-ended DAC outputs are used, then a fixed Vref becomes necessary, which may come from a 3rd DAC, or from the VREF of the signal DACs. Vref is simply the zero reference of the DACs, and it can be shared between I and Q channels, but make sure it is low source impedance, for minimum crosstalk between I and Q channels. Buffer VREF with a LTC6246 voltage follower if necessary.
In this example the single-ended I and Q DACs are configured for 0 to +1.024V output. Note that the DAC digital drive should be scaled to use all available DAC bits, but never touch the full scale extremes.
Optional resistors R5 and R6 are chosen for a low average DC current from the DACs. In this example, the input common-mode voltage is pulled down to 0.512V to match the center of the DAC output voltage range.
Single-Ended to Differential Driver
Using differential baseband signals to drive the modulator is highly preferred, as opposed to single-ended baseband drive. Differential drive offers 6dB higher RF output power, and the lowest error vector magnitude (EVM).
Choosing an amplifier with differential input and output also simplifies the design, because
- The design symmetry reduces output DC offset error, which reduces LO feedthrough (LOFT) and improves modulation accuracy.
- Output common mode voltage (VOCM) can be changed independently to meet the requirement of the modulator, and
- Gain can be changed independently without affecting VOCM.
Figure 2. A differential driver simplifies the design by offering independent adjustment of gain and output common-mode voltage.
The LTC6362 is selected for this application because it draws only 1mA of supply current, and noise is quite low. Low amplifier noise is important for designs where the baseband signal source has low close-in noise density, which means a very high order baseband digital filter is being utilized, along with high resolution DACs. We don’t want the amplifier to make this close-in S/N ratio any worse.
Capacitors C10 and C11, are chosen to curtail HF noise while providing negligible (<0.1 dB) amplitude error contribution at the highest baseband frequency. The LTspice simulation is quite useful for this purpose.
DAC LC Re-Construct Filter
The DAC re-construct filter is important for smoothing out the DAC stair-step output, which reduces the high frequency Nyquist images. Another important function is to minimize the wideband noise floor at the modulator output.
Figure 3. LC Re-construct Filter. This design is for 266 Ohm source and load terminations.
Use classical filter design CAD tools to synthesize these LC element values for your design. For best modulation accuracy, choose Bessel LPF response for linear phase. In this example, the -3dB corner is placed at 50 kHz, which is good for symbol rates up to ~30 ksps.
The amplitude of the DAC image spurs, relative to the desired signal, can be estimated by SIN(x) / x, where x = π f / fCLK, For the frequency plan of this design we can expect 25dB image attenuation at 220 kHz. Add to this the attenuation afforded by the DAC LC reconstruct filter at 220 kHz (45dB), in order to estimate the total Nyquist image rejection of 70dB.
Deriving the VOCM Source
When VCC is well regulated, VOCM can be derived from a simple Vcc resistor divider. Otherwise, VOCM is efficiently derived from an adjustable LT3009 voltage regulator, which supports resistor divider current as low as 1µA. Proper VOCM voltage is essential for establishing correct IQ modulator operating point.
Figure 4. Simple LT3009 provides well regulated, low-noise Vocm, with minimal current waste.
Note the LTC6362’s have an internal VOCM voltage divider at VCC/2, so whenever Vcc exceeds 2.8V, VOCM needs some pull-down to stay at 1.4V. This is the purpose of R3.
Typically, for many LTC5589 and LTC5599 applications, a differential drive level of approx. 330mVRMS is a good starting point, measured differentially at the I and Q inputs to the IQ Modulator. Because of LC filter losses, the drive to the input of the LC filter is typically slightly higher, around 358mVRMS at each diff. amp output. A spectrum analyzer should be used to confirm that the desired modulator RF output spectral purity is achieved.
Drive the IQ Modulator with higher VRMS for higher output power, or lower VRMS for lower IMD and better modulation accuracy. The amplifier feedback resistors may be adjusted for this purpose.
The filter synthesis should already be correct. Yet because the filter design must be converted from single-ended to differential, there is some room for design error to creep in. Another source of error can be delay in the amplifiers, which becomes a factor for higher symbol rates. A quick LTspice simulation can be helpful to verify that both of these sources of error remain benign.
Begin by looking at the passband and stopband responses separately:
Figure 5. Overall circuit passband response. Here we are looking for minimal gain roll-off and group delay variation at the highest signal frequency (20.25 kHz in this design example).
Note the response droop at the highest signal frequency of 20.25 kHz is only 0.5dB. Below 1dB is a good goal. And group delay is so flat that you’ll need delta-marker to measure it on the simulation graph above. It is only 3.7ns, a negligible amount compared to a symbol period of 1/30k, or 33.3us. Less than a few percent of a symbol period is a good goal. Group delay variation is very good in this case, thanks to the Bessel filter.
Figure 6. Overall circuit stopband response. Here we are looking for sufficient attenuation at the Nyquist Image frequency (220kHz in this design example).
In this example, we see 46dB attenuation out at the image frequency. Combined with the 25dB sin(x) / x, that’s 71 dB. A good goal here is around 70dB or better. For over-the-air applications, regulatory agency requirements come into play here.
Note that the filter element losses can be not so easy to simulate accurately. Their effect will be evident in the next subject:
For digital modulation, the ultimate test of modulator accuracy is EVM.
Figure 7. LTC5599 EVM measurement of 0.8% rms aligns with expectations. All SPI registers at default values except for LO frequency which is set at 450MHz.
Considering that this system is utilizing 8-bit DACs, it’s no surprise that the EVM measures around 0.8%, because fundamentally that’s about the best attainable, considering that the ENOB is going to be less than 8-bits:
Figure 8. Ask for your free Linear Technology Accuracy Translator! In this case it shows how the effective DAC resolution can limit modulation accuracy.
EVM test results align with previous published measurements and are definitely good enough for most digital modulation applications.
Further EVM improvements would require more FIR filter taps, higher resolution DACs, and perhaps some digital filter compensation for both the sin(x) / x roll-off and the DAC re-construct filter roll-off.
Figure 9. RF Output Spectrum at -4dBm output power. Note the worst DAC image is about -70dBc, good enough for most applications.
Looking at the output spectrum, the close-in noise spectral density is determined by several factors: the FIR filter order, LO phase noise, and modulator drive level.
This blog illustrates the detailed design techniques for utilizing low-power IQ Modulators such as the LTC5599 and LTC5589. For highest modulation accuracy, lowest noise, and lowest DC power consumption, the designer must carefully consider each circuit block, as presented here by way of example.