Bringing NMR into the Digital Age
PT2026: What's so hard ?

Metrolab’s new generation, all digital NMR teslameters promised dramatic improvements: ten times the frequency range, six times the measurement rate, three times the acceptable field inhomogeneity, four times the tracking rate, two channels instead of one, color touch screen, USB and Ethernet interfaces - to name just a few. But the release of the PT2026 has been pushed back time and time again. Why - what’s so hard? For the technically minded, here’s a look under the hood.
 


Initially, the PT2026 development was delayed by “normal” problems, of the type known to every engineer: overheating, interface bugs, optimization of the RF circuitry, underestimated difficulties in firmware development, etc. Once all those were resolved, we were finally ready to discover the real challenges.
But to explain these, we have to review the operating principles of the previous-generation, analog NMR magnetometer - the PT2025 - and its digital descendent, the PT2026.

The PT2025 (show architecture diagram) is designed around a high-precision Voltage Controlled Oscillator (VCO). In the Manual mode, the VCO input voltage, and therefore the frequency, is set by front-panel controls. This RF frequency is divided down to match the probe range, and sent to a coil surrounding the NMR sample.
A separate oscillator and coil induce a slight modulation of the magnetic field at the NMR sample; this causes the NMR resonant frequency of the sample to oscillate. When it crosses the RF frequency, the resonance is detected and amplified with an ultra sensitive peak detector, located right in the probe.
Concretely, the peak detector detects a very slight dip in the RF signal, after it has been filtered by a finely tuned LC circuit formed by the RF coil and a varicap.

From analogue to digital, with the same probe
Still in the PT2025, this low frequency, peak-detected signal – the “NMR signal” – is amplified and sent back to the main unit, where it serves to drive the auto-tuning circuitry for the LC filter and, more importantly, to indicate that the NMR signal has been found. The strength of the magnetic field corresponds to the RF frequency, measured by a high-precision counter. Finally, in the Auto mode, an additional feedback loop automatically adjusts the VCO input voltage to keep the NMR signal centered within the field-modulation sweep.

The PT2026 (show architecture diagram) can use the same probes, but has a much simpler architecture. The RF signal is generated using a Direct Digital Synthesizer (DDS). This allows the frequency to be rapidly swept up and down (chirping), thus replacing the field modulation with a frequency modulation and rendering the field modulation coil unnecessary. The NMR signal is digitized and fed into a DSP, thus allowing much more sophisticated detection algorithms than a simple threshold. The DSP also controls the DDS, thus guaranteeing that the DSP knows exactly what part of the signal corresponds to what frequency. Finally, the tuning is also controlled digitally, to match the RF frequency at all times.

Two flies in the ointment: one fly removed…
A number of factors kept the DSP from properly tuning the LC circuit in the probe. First, the tuning frequency was hypersensitive to variations in DAC value, resulting in a very touchy control system. In addition, the existing probes have large time constants that prevent us from tracking the frequency during the frequency modulation cycle. Finally, the peak detector acts as a differentiator, resulting in a sloping baseline.

 


It took months to fully understand these problems and explore hardware and/or software solutions for each of them. The final solution has two parts: revert to an analog autotuning circuit, and use a baseline extraction algorithm in the DSP. The new analog regulator is actually a good thing; thanks to its clever, minimalist design it can be implemented in the probe itself, allowing rapid tuning and fewer constraints on the probe cable length. For compatibility with existing probes, the same circuit needs to be replicated in the PT2026 main unit.

…and the fly that wouldn’t die
The other problem is that a DDS approximates a sine wave with a series of linear segments. As with any sampled system, this generates harmonics of the DDS clock that are “folded down” into the operative frequency range and, once detected, appear as spurious NMR signals. The PT2026 included a mechanism to change the DDS clock frequency to “move” these spurious signals “out of the way”; however, we completely underestimated their number and perniciousness. A scan through the frequency range of a single probe shows that the NMR signal (red arrow) is completely drowned out by spurious signals.


After months of work, clever signal processing finally allowed us to make a reliable distinction between spurious and real NMR signals – assuming the real NMR signal didn’t happen to fall on the same frequency as a spurious one! No amount of cleverness helps in that case. In fact, we found that this is exactly what happens, and our probes had a number of “dead spots” where they could not measure. This was a compromise we were not willing to make, so we finally had to admit that this approach was a dead end.

So now what?
It has become clear that the PT2026 hardware needs to be changed to work with existing probes. Currently we are testing an approach that combines the clean signal of a VCO with the absolute frequency control of a DDS. Simulations have been very promising, and we are now building a prototype – stay tuned! (so to speak…)
For the longer term, we have demonstrated that the existing PT2026 hardware works well for pulsed-wave probes (another approach to detecting NMR resonance – see our NMR Technology page). Pulsed wave detection is much less sensitive to tuning, plus it allows an NMR signal to be distinguished from a spurious signal even if the two have the same frequency. However, this approach breaks all compatibility with existing Metrolab probes, and thus marks a radical new departure.

Downloads:
PT2026 data sheet