SSB/CW/FM signal generator 35 - 4400MHz

Author: Georg Seegerer (DG6RS)

email: dg6rs (at) t-online (point) de

Deutsche Version dieser Seite

Summary

The signal source works on 35-4400MHz and produces various narrow band modulation schemes (2.7kHz maximum) via polar modulation: SSB/CW/FM/FT8/PSK31/etc. The concept to produce the signal via combination of amplitude and phase modulation is similar to the uSDX by DL2MAN. However the power amplifier is not part of the signal source. Output power is approx. 0dBm. Intention was to build a signal source for the narrow band transponder of the QO-100 satellite, without the need of a 2m/70cm allmode transceiver as would be necessary for a usual upconverter.

I plan to publish hard- and software as open source. It is planned to release kits with the PCB and pre-soldered SMD components. To stay informed, visit this website regularly or write me an email to receive updates: dg6rs (at) t-online (point) de

As a first step the hardware was built from readily available PCBs. Now a first version of an all-in-one PCB exists. On the left side the low-frequency signals are connected (microphone, PTT, RS232, etc.). On the right side there is the SMA connector for the SSB signal on 2.4GHz. The unpopulated SMD components on the right side are for the mixer of a downconverter (RX path, not used for QO-100 uplink).

The software is available on gitlab now.

Start of the Project

The idea was born when I intended to build a ground station for the QO-100 satellite but to develop at least one component on my own. A survey of available radio frequency components on Aliexpress was another starting point of the project. A polar modulator generates the complex modulation in the IQ plane using phase and amplitude instead of the I and Q coordinate. It is the basic idea of the uSDX from DL2MAN [3] but at a frequency in the UHF/SHF range.

During the development phase of this signal generator only a cheap RTL-SDR stick was necessary to listen to the audio of the modulation. Only later in the project a professional spectrum analyzer was necessary to measure the ACPR (adjacent channel power ratio).

The usual way to generate SSB

Usually a SSB signal on 2.4GHz band is generated by using a 2m/70cm allmode transceiver with an upconverter to 13cm. Of course the signal generator needs to generate the SSB signal in the first place. Two methods are widespread to be used: First, the filter method with a narrowband quarz filter at 10.7MHz which is narrow enough to separate lower and upper sideband. The second method is to use a hilbert transformer for the audio signal to make a wideband phase shift by 90°. Then an IQ-modulator can shift the signal to the intended transmit frequency.

An IQ modulator consists of two frequency mixers which are driven by the same LO, but one mixer gets the LO with a 90° phase shift. At this point the phase shift does not need to be broadband, therefore no hilbert transformer is necessary. IQ modulators usually are manufactured on one chip which contains both mixers and the 90° phase shifter because equal gain and delay on both paths is critical for a good suppression of the unwanted side band. Usual values are 30-50dB.

A good source for further information about various methods of SSB generation can be found in [1]. About the Weaver method [2] (text in german).

Polar Modulator for SSB

The necessary trajectory in the IQ diagram also can be generated by modulating the amplitude and phase, not the I and Q component. The phase modulation is done by the ADF4351 which gets a frequency set point up to +/-6kHz from the nominal frequency. A digital attenuator does the amplitude modulation. It is set from 0 to 31.5dB in steps of 0.5dB.

The FM and AM signal is calculated by a STM32F103 with an update rate of 16kHz. The algorithm uses only add, bitshift and multiplication of 32 bit integer numbers. No further DSP functions are necessary. The CPU runs with 72MHz and has a usage level of 80%.

One advantage of the polar modulation lies in the fact that there is no LO (local oscillator) and no mirror frequency which need to be filtered. This makes it easier to develop hardware which covers several HAM radio bands. Usually only a low pass filter is necessary for each band.

The biggest disadvantage of polar modulation is its limitation to rather narrowband signals. This project here covers the 2.7kHz bandwidth of a speech SSB channel. DATV (digital amateur television) with 2MHz bandwidth clearly is outside of the possible or reasonable bandwidth of a polar modulator.

The transformation of I and Q to amplitude/phase signals is a nonlinear function. For correct representation of the output signal, the amplitude and phase signals need an increased bandwidth compared to the output signal. Rule of thumb is a 5 times increase of bandwidth. The recombination of amplitude and phase need to properly work in this increased frequency range to ensure no ghost signal appear outside the bandwidth of the SSB signal. A non-linearity in the frequency response of the amplitude or phase channel leads to deviations which mainly cause ghost signals near the SSB signal, i.e. cause a degraded ACRP (Adjacent Channel Power Ratio).

Usually the reason to use polar modulation is to combine the amplitude and phase signal in the power amplifier. This enables to use a class C/E/F nonlinear amplifier with high efficiency for a linear signal and still get a non-distorted signal.

In this project, this is not used (until now). The separation and re-combination of amplitude and phase signal is merely done because the separation of the signal generation into FM (with a PLL/VCO chip) and AM (with a digital attenuator) makes it possible to use wide spread chips which are produced by several manufaturers (second source). IQ-mixer-chips usually are single-source, and curretn models are optimized for signals with bandwidths of several MHz. Additionally, with a polar modulator there is no LO which need to be shielded properly against the PA.

Hilbert Algorithm

The usual hilbert algorithm does not calculate the Q-component for a given I-signal as you might assumed. It is easier to use the (real) input signal and calculate two output signals which have a frequency-dependent phase shift, but a differential phase in between the signals of almost constant 90°. Two all-pass filters with 8th order were implemented. Coefficients are from the paper [9].

Every second coefficient is zero, i.e. 4 multiplications are needed per filter per audio sample. The coefficients are calculated for 44.1kHz sample rate and the filter works for 20-20000Hz. Used with 16kHz sample rate the frequency band with 90° phase difference at the output ranges from 7.2Hz to 7.2kHz. This seems to be overdone, but the next smaller filter I found had 4 pole all pass filters and had the 90° phase difference from 300-3000Hz. However, the unused sideband only gets suppressed in the frequency band in which the hilbert algorithm outputs its 90° phase difference. Thus, it is necessary to have an hilbert algorithm which works for the whole frequency range where you might have an audio signal, not only in the frequency range you consider the used range. Therefore, the transition area of the audio filters between passband and stopband needs to be covered, too.

Cordic Algorithm

The Cordic algorithm (Coordinate Rotation Digital Computer) calculates the transition from cartesian (X/Y) to polar coordinates (amplitude/phase). The algorithm works purely with integer add and bitshift, therefore is suitable for CPUs without math coprocessor. There are variations of the algorithm for all trigonometric functions. An introduction can be found in [10].

Adjustable Reference Frequency

A fractional-N-PLL theoretically can produce any frequency, the resolution only limited by the denominator of the fractional counter. The ADF4351 has 12 bit registers for the nominator and denominator. In the case of a phase frequency comparator frequency of 1MHz, this calculates to a frequency step of 244Hz.

The reference frequency needs to be slightly shifted due to two reasons. First, a fractional-N-PLL generates an elevated level of spurs in the case the fractional part is very small, i.e. the frequency generated is near a frequency step of a integer-N-PLL. Assume a PFD (phase frequency detector) frequency and using the signal generator for QO-100, these frequency points are 2400 and 2401MHz. As 2400MHz is used for a beacon which you need to leave at least 5kHz gap, you can ignore this issue for QO-100. If you want to use the device as general signal generator for SSB or FM, e.g. in the 70cm band you would encounter such a frequency point every 125kHz (i.e. 430MHz, 430125kHz, 430250kHz, etc.), because the internal VCO (2200-4400MHz) is divided by 8 in this frequency range. This would be an unacceptable restriction for choosing the TX frequency. A second reason is that due to the phase modulation in SSB mode the VCO gets a FM up to +/-6kHz around the nominal frequency, so even more distance needs to be kept from the critical frequency points.

For these reasons the reference frequency of the ADF4351 is made adjustable, so for every TX frequency the divider in the PLL can be set up in a way that the fractional-N-PLL works roughly in the middle between two interger-PLL frequency steps. For a 10MHz reference, a maximum frequency shift of approximately 2kHz is necessary.

Initially, I intended to use a Si5351 frequency generator for the adjustable reference frequency. However, it turned out that the chip generates a vibrato, ca. 300Hz frequency deviation at 1242MHz (corresponds to 2.5Hz at 10MHz reference frequency), and with ca. 3Hz repetition rate. I don't know whether this effect can be avoided by a better configuration. Several configuration attempts of the PLL stages did not change anything about the vibrato (fractional or integer divider of the PLLs). At other modulation types this small and slow vibrato would not be noticed. With FM, the 3Hz repetition rate is below the audible frequency range. For digital modulation types, the receiver would compensate for the frequency or phase deviation.
The effect, recorded with a SDR (Video, 11s, 2MByte)

Therefore, a VCXO (voltage controlled crystal oscillator) need to be used to generate the adjustable reference frequency. However, is not implemented yet.

The Prototype

The PLL and the digital attenuator get data simultaneously via two SPI buses. The signal for chip select is the same for both chips and is controlled via software. Usually this signal is named CS (chip select). However, on the used PCBs this signal is named LE (latch enable). Please note that the ADF4351 has an input named CE (chip enable) which is a power down function and has nothing to do with the SPI bus. The CE input of the ADF4351 needs to be high for the chip to be active.
Further connections of the ADF4351 are LD (Lock Detect) and MUX (multiplexer output).

The Si5351 generates the reference frequency for the ADF4351 and is controlled via I2C bus.

The control of the TX parameters (TX frequency, power level, etc.) is done via the RS232 connection. The concept allows the control panel and the RF signal generator being separated. The connection cable in between transfers RS232 (115.2kBaud, bidirectional), PTT key and analogue audio. This is 6 wires if the audio is transferred symmetrical. Audio can be streamed via RS232, but only with ca. 11kSps and 8 bit resolution.

Measurements

Most urgent question: How does the modulation sound? The SSB signal was generated in the 23cm band, the highest HAM radio band which I can receive with the RTL-SDR stick. In the video the out-of-band signals in the neighboring channel can be seen which are caused by non-linearities.

CQ call (Video, 12s, 5MByte)

Audio from a German speaking number station (Video, 14s, 6MByte)

A waterfall diagram from a transmission via QO-100. A bit of signal in the LSB can be seen. Apart from this, no out-of-band signals which are strong enough to be seen. Audio sounds a bit dull, and in the waterfall it can be seen that the possible bandwidth of 2.7kHz of a SSB channel is not completely used. So, more fine tuning of the signal path is necessary.
Waterfall diagram of QO-100 transmission (Video, 11s, 1MByte)

Similar Projects

Another project which works with polar modulation is uSDX from DL2MAN [3]. He uses the amplitude signal to drive a non-linear PA to get good efficiency. The uSDX works in the short wave range from 80m to 20m.

Other projects use IQ modulators to generate a SSB signal. Quite some projects are designed for 13cm, mainly because of the QO-100 satellite. Usually the audio signal is fed into a hilbert transformer which generates inphase- and quadrature component. These signals go into two DACs into the analogue domain and then to a IQ modulator chip.

The modulator of PA0RWE [4] uses an ADF5375 as IQ-modulator for 400-6000MHz, an ADF4351 for the LO and a dsPIC as digital signal processor for the IQ signal.

The QO-100 cube [5] from OH2UDS/TA7W uses an ADF4360 to generate the LO and an AD8346 as IQ-modulator for 800-2500MHz. The base band signal is calculated by a RP2040 (dual core ARM CPU). The project is a complete full duplex QO-100 transceiver with a water fall diagram in the receiver. It is currently under development and had a presentation on the 2024 HAM-radio festival in Friedrichshafen, Germany [6].

These three projects only serve as a examples.

Why use polar modulation? And why not?

Why polar modulation is quite unknown? In commercial applications, most commonly ASK or FM is used for narrowband signals. Broadband signals in the GHz range typically have a bandwith of 5MHz (LTE) or 20 to 80MHz (WiFi). This is the reason why attempts to increase the efficiency of power amplifiers with polar modulation disappeared in the early 2000s, mainly with the transition from EDGE (2.5G, bandwidth 200kHz) to UMTS (3G, bandwidth 5MHz). Additionally, as linear modulation schemes became more widespread used in portable devices, more effort was put into optimizing the efficiency of linear amplifiers. Therefore, the increase in efficiency to be expected from polar modulation became smaller.

For stationary devices using small TX power, like base stations for cellular radio, the efficiency of the PA was never a big topic because other components like digital processing or air conditioning dominate the power consumption.

In cases with high TX power and small bandwidths, e.g. medium and short wave broadcast transmitters, polar modulation was always employed. For pure AM, it was not called polar modulation, it was rather modulation of the anode voltage. The modulation got polar when the transmitter was used for „Digital Radio Mondiale“ (DRM), that is an OFDM modulation (Orthogonal Frequency-Division Multiplexing) with 9kHz bandwidth. The signal generator was used for the phase part of the modulation, and the AM was done by modulating the anode voltage as usual.

HAM radio operators use a 2.7kHz wide SSB signal even at frequencies in the GHz range. This is a narrowband signal which is possible to be generated with polar modulation. In this application polar modulation can find a niche.

Possible and/or planned further developments

Hybrid Converter

Until now, the SSB signal source is a lab prototype made with readily available PCBs from Aliexpress. Next step will be to make a layout to have all components on a 4-layer PCB. Doing this, it would be a small extra effort to add the functionality of a receive path. It's only a few extra components. In the RX direction the function would be a usual downconverter which converts the received signal from the 23/13/9cm HAM radio band into the 70cm band (or 2m/10m). Then it can be demodulated by a 70cm receiver or a RTL-SDR stick. The SDR has the advantage of a water fall diagram, so you don't have to tune over the band just to see which frequencies are occupied. On the transmit side, the signal is completely generated by the module, no allmode transceiver necessary.

User Interface

Currently, the signal generator is controlled via ASCII commands on RS232, so it is possible to control the generator with a terminal program on a PC. In the future there may be a hardware user interface with the look-and-feel of a two-way radio. It will be possible to have a distance between the user interface and the RF signal generation, this is necessary on quite a few QO-100 stations. Cable for RS-232 signals and analogue audio are cheaper than for 2.4GHz RF signal.

CESSB (Controlled-envelope SSB)

For SSB there is always the attempt to increase the mean power while keeping the peak power constant. The wikipedia article [8] about CESSB (Controlled-envelope SSB) states that for speech transmission the peak-to-average ratio can be reduced by 3.8 dB:
„The standard SSB envelope peaks are due to truncation of the spectrum and nonlinear phase distortion from the approximation errors of the practical implementation of the required Hilbert transform. It was recently shown that suitable overshoot compensation (so-called controlled-envelope single-sideband modulation or CESSB) achieves about 3.8 dB of peak reduction for speech transmission.“

As the audio signal gets processed digitally anyway, such methods can be used in the signal generator. However, until now implementation did not start.

AM/PM for frequency multiplier

For transmissions in the two-digit GHz range the TX frequency often gets generated via a frequency multiplier chain. Its available RF power can be directly used for CW. However, for SSB this RF power is fed into an mixer as LO, the available transmission power is about 10dB lower.

A frequency multiplier chain broadens the input phase modulation by the multiplication factor. Amplitude modulation does not have this effect, but is probably severely distorted.

It should be possible to (partly) compensate the nonlinear AM reaction of the multiplier chain by predistortion. The phase deviation is reduced by the factor of frequency multiplication. The ADF4351 can produce the RF signal up to 4400MHz which can help to reduce the number of multiplier stages.

Using this method it should be possible to generate a SSB signal with the same peak power as the CW signal.

A test of the theory is still missing. I would be delighted about contact to microwave enthusiasts.

Driving non-linear PA

Similar to the uSDX project, it would be obvious to drive a non-linear PA (class C/E/F) with the amplitude signal. This would necessitate to develop a PA specially for the use with this signal generator. The advantage of a broadband generator would be lost, as a PA module tends to serve only one HAM radio band.

In case of a contact who is interested in the development of such a PA, I can do necessary changes to the software, like having the amplitude signal on the SPI bus for a DAC or as PWM signal.

Series Production

I look forward to questions and discussions about the project. The number of contacts will decide whether the signal generator stays a single-of-its-kind for my QO-100 station or will be available as PCB assembly for series production.

Hard- and Software will be available as open source soon.

Contact: Georg Seegerer / dg6rs (at) t-online (point) de

Acknowledgments

The uSDX from DL2MAN [3] was the inspiration to start the project in the first place.

Information and demo code from Hans Summers and QRP Labs helped a lot to get the Si5351 running.
https://qrp-labs.com/synth/si5351ademo.html
http://www.hanssummers.com/

The fablab of unversity Erlangen and Jan DL3KBF helped me with RF measurement equipment.

My wife tolerated all-night programming, soldering and testing.

List of references

List of Abbreviations