A GPS Disciplined OCXO Frequency Standard & NTP Reference Clock
The Quartzlock E8000 represents a breakthrough in low cost, 1U rackmount, traceable, calibration-free "off air" frequency and time standards. It is a GPS controlled frequency and timing source, with both 10 MHz frequency output and 1 PPS timing mark synchronized to GPS time. Features include a holdover function on both the timing (1 PPS) and the frequency output. The module may be configured with a variety of controlled oscillators, from low cost OCXOs to rubidium standards, with the short-term stability (Allan variance) and holdover performance being a function of the controlled oscillator. The 10 MHz output phase noise specification is -110 dBc/Hz at 1 Hz offset.
A key component of the E8000 is a special GPS receiver module, which has enhanced performance for timing applications. This is significant as the choice of GPS receiver is vital in the design of a GPS controlled frequency/time standard. All GPS receivers make a calculation of GPS time every update interval, often one second. However, methods of using this information, and communicating it to the outside world, vary.
The GPS receiver chosen for the E8000 has a programmable frequency output. This is a square wave generated from an NCO (a simple version of a DDS without the DAC) clocked from the internal 120 MHz clock. Every occasion that the time solution is calculated, the phase of the NCO is adjusted so that the frequency output is accurate. This frequency output means that a phase lock loop can be used to lock the controlled oscillator to GPS time.
Time Solution Accuracy
A further consideration in the choice of GPS receiver is the accuracy of the time solution. A conventional navigation receiver will have a typical variation in its timing calculation of 0.5 to 2 µs RMS. This is a random variation caused by noise and ionospheric effects. There will also be a systematic longer term (minutes to hours) variation due to the satellite constellation continuously changing.
A navigation receiver needs at least four satellites to calculate a 3D position and time. More satellites will add redundancy, and will thus reduce errors in the position and time calculation. A special timing receiver assumes that the receiver is stationary; therefore, the position does not change. The pseudo range measurement from each satellite is only used to calculate the time, giving greater redundancy and less jitter.
If the position is not correct, then the 1 PPS time output will have a systematic error. In order to avoid this, a timing receiver will have a self-survey mode where the position is calculated and averaged continuously for several minutes. After the self-survey is complete, the receiver switches into position hold mode.
A timing receiver operating in position hold mode will have an RMS jitter on the GPS time calculation of between 5 and 50 ns RMS. The GPS receiver chosen for Quartzlock designs, including the E8000, has a measured RMS jitter on the outputs of about 7 ns.
A GPS receiver used for timing applications will often have a built in monitor of the likely timing accuracy. The algorithm used was developed by Motorola, and is called the Time-Receiver Autonomous Integrity Monitoring (TRAIM) algorithm. If the predicted error in the time calculation increases above a threshold, default 300 ns, the GPS receiver will go into holdover mode. In this mode the 1 PPS output time is held, and the frequency output is locked at its current frequency. The GPS receiver will also set its digital output to indicate that the receiver is in holdover. The microcontroller can then freeze the tuning voltage of the controlled oscillator, putting this into holdover mode.
The drawback with this simple design is that the GPS receiver uses a low quality internal TCXO for its internal clock. When the GPS receiver goes into holdover, it freezes the output time of the 1 PPS. However, as the 1 PPS is timed internally from the TCXO, it will drift and will not give very good holdover performance. What is required is a 1 PPS time pulse that is timed from the controlled oscillator. In this way the holdover performance of the 1 PPS will be identical to that of the controlled oscillator.
If the controlled oscillator is a rubidium standard (E8010), then the holdover performance will be 1 µs a day. The solution is to use the GPS receiver in external clock mode. If this is done when the GPS receiver goes into holdover, the 1 PPS timing will come from the controlled oscillator.
The hardware and software used to lock the controlled oscillator to GPS time is a classic PLL implemented digitally using a microcontroller. The frequency output from the GPS receiver is set to 2.5 MHz. This is then compared with divided outputs, also at 2.5 MHz, from the controlled oscillator using a quadrature phase detector. The microcontroller then samples the outputs from the phase detectors and implements the PLL.
This PLL is slightly unusual in that the controlled oscillator has an effect on both inputs of the phase detector. On one input the controlled oscillator is divided down to 2.5 MHz and is applied directly to the phase detector. On the other input the 2.5 MHz comes from the GPS receiver, which is clocked from the controlled oscillator multiplied to 20 MHz. The effect of the GPS lock may be considered as a variable divider, which sets the frequency output to 2.5 MHz regardless of the frequency of the controlled oscillator.
However, there will be regular phase adjustments. These occur because the binary NCO in the GPS receiver cannot divide by the exact decimal divider required to generate 2.5 from 120 MHz (the GPS receiver internal clock). These regular phase adjustments are filtered out by the narrow bandwidth of the digital PLL.
A further effect is that the response of the GPS receiver to a change in frequency of its clock is not instantaneous. There is a delay between a shift in clock frequency and the restoration of the frequency output to its nominal value. The effect of this delay is to add a further time constant to the PLL loop filter. This unavoidable extra filter means that PLL bandwidths cannot be too wide without loop instability developing.
For the E8000, the time from switch-on to the frequency standard becoming useable is a function of many things. These include the GPS receiver cold start, self-survey time, reference oscillator warm-up, and PLL locking and stabilization time. Sometimes a switch-on will follow a change in position of the receiver, but more often the position will remain unchanged. In spite of this, for design and user simplicity, it is assumed that the receiver has moved after every switch-on. Thus, the GPS receiver will carry out its self-survey after every switch-on. This takes about 15 minutes.
Most controlled oscillators are very far from their final frequency during warm-up. An OCXO may be 10 ppm in error, and a rubidium will sweep its OCXO during warm-up in order to find the atomic resonance. The GPS receiver will not operate properly if its external clock is not stable to within 1 ppm. For this reason the GPS receiver is held in reset until the controlled oscillator has warmed up.
At this point the PLL is activated and starts locking. As the tuning voltage for the controlled oscillator is stored in EEPROM from the last time the PLL was in lock, the frequency error at the phase detector should be very small. However, there may be an initial phase error of up to 2π radians. This applies when the phase/frequency detector is in use, which it is during initial locking. Correction of the initial phase error can only be made by offsetting the frequency of the controlled oscillator. This is quite normal for any PLL. In order to minimize the phase pull in time the PLL bandwidth is set to a maximum during locking.
Figure 1 38000 With high quality OCXO, reference passive H maser.
The E8000 has been tested with a variety of controlled oscillators. For all the tests the GPS antenna was on a flat roof with an unobstructed view of the sky. The Allan variances (AVAR) can be seen in Figures 1 (OCXO) and 2 (rubidium). With a rubidium oscillator, the measured Allan variance was below 1X10-12 at averaging times longer than 100 seconds, with a floor of about 4x10-13. With a high quality OCXO the Allan variance was about 1x10-11 between two seconds and 100 seconds, and fell to less than 1x10-2 at 10,000 seconds.
Figure 2 E8ooo with rubidium, reference passive H maser.
The E8000 now has a standard performance level of -110 dBc/Hz at 1 Hz offset for phase noise and a short-term stability of 8x10-13/s. Optional -115 dBc/Hz at 1 Hz offset and AVAR of 6x10-13/s are also available. This performance and choice in a low cost GPS time and frequency reference is a major breakthrough. The E8000 has full NTP server plus separate RS232 for GPS View/WinOncore 12.