I downloaded the logs for the Chinese xxx mAh battery and the old Li cell from an Apple laptop battery pack. Importing those logs into excel and extending the data now gives me:
No evidence of damage after 650 charge/discharge cycles. Similarly, the cell recovered from an Apple laptop pack shows no sign of deterioration after some 270 cycles.
Maybe there is something in pulse charging? I'll just let everything keep running to see what happens.
The bridge is the real point where the VNA measures the reflected signal from the load connected to Port 1. It is inside the VNA. Calibration is a mathematical process used to move the apparent measuring point to wherever the OSL standards are connected during calibration.
What happens when the piece of coax connecting the VNA to the calibration point is not 50ohms? The 50ohm load gets transformed to a different value depending on the electrical length of the coax at the measurement frequency and the coax's impedance. The implicit assumption is that the VNA can measure this transformed load with sufficient accuracy that the error in subsequent measurements can be ignored.
But can it? Does this implicit assumption fail as the reflected signal level falls, ie the connected load approaches 50R? And a what point does it fail? Here is how I made my own mind up. I used the best of the 4 VNA's I have for this, a NanoVNA V2 plus4. The one in the metal box.
Step 1 : Perform an OSL calibration directly at the SMA connector on the VNA. The calibration load is a string of SMA attenuators with the 50R termination at the end of them.
Step 2 : Plot S11 against frequency. Also understood to be the input return loss by most people. Save that to one of the memories.
Step 3 : Remove the 50R load but not the string of attenuators. You are now measuring an input return loss twice that of the connected loss in the string of attenuators. In my case, with 45dB of attenuation in the string, that should be a return loss of 90dB. Save to another memory
Step 4 : Repeat Step 3 but remove the last attenuator. I now had 35dB of attenuation which should show as a 70db return loss.
Step 5 : Remove the next attenuator. I now had 25dB of attenuator which should show as a 50Db return loss.
Step 5 : Remove the next attenuator. I now had 20dB of attenuator which should show as a 40Db return loss.
So what conclusions can we draw about this NanoVNA at this point?
Firstly, it's not very good at measuring loads which have a return loss of under 60dB. Instead of flat lines we have odd looking curves.
Secondly, when the return loss is 40Db or better I'd be circumspect in the actual displayed value.
Nevertheless, a useful instrument since a return loss measurement down to 40Db is useful.
Now, lets attach a piece of coax and reassemble the attenuator string for 25dB of attenuation. If the piece of coax had a characteristic impedance of 50ohm this should show as a 50Db return loss. Clearly, from the measured result, the nice blue cable is not 50ohm.
 |
| Should be the same line as the 50dB curve in the chart above. |
At the device the first graph shows a relatively flat line. With the piece of coax the graph suggests the coax is not 50ohm.
Now, repeating the OSL calibration at the end of the coax, which in practice is what I would do, the 25Db attenuator string still returns an error prone measurement though a 15dB attenuator appears satisfactory up to 800MHz.
For the price the NanoVNA is still the most useful piece of test gear I own and I recommend you have one. However, be aware of the limitations. When measuring S11, or input return loss, the value returned when better than 30dB is at best indicative. Careful calibration cannot overcome this.
You can test, in relative terms, how good the cable you are using is by performing the calibration directly at the device then moving the reference load to the end of the cable and measuring S11, or input return loss. I've tested scores of cables and none of them are satisfactory. But it allowed me to weed out some truly terrible cables.
Finally, there are a number of posts I have read that go through some complicated mathematics, typically using python, to derive some improved number or esoteric measurement. Must make the author feel clever but they probably don't appreciate the flaws of the measurement underpinning whatever they are trying to do. The pitfalls of garbage in means garbage out. And no amount of ridiculing others by referring to their work as "Hammy Sammy" make what they do superior.
Update 15/09/2024 :
The 14Ah cells were replaced by 32AH cells and the difference was almost frightening. Despite having been in hospital for 6 weeks the mower starts within a second of cranking. There is no sag in the crank speed, it's a case of turn key and let it go.
The BMS still performs with aplomb though the dump resistor burnt out. I replaced it with a 55W headlight globe. Still Chinese watts though so who knows if it will last.
The system has proven to be robust and reliable. The only downside is I can no longer avoid mowing the lawn due to a flat battery!
If you're interested in more details you're welcome to contact me.Not much to report at present because a recent and unexpected diagnosis of Acute Myeloid Leukemia meant a plus 5 week hospital stay undergoing further testing and chemotherapy.
I did however download the log of test results before my stay for one of those chinese 18650 xxxxx mah cells. This was a new, but old stock, battery and had been used in a pulse charger to test the hardware and software. After 322 discharge cycles there did not appear to be any degradation in reported capacity.
A few hours at home yesterday on day release meant I could download the logs for the abused tablet Li-Ion batteries. They too showed no signs of degradation after large number of cycles.
I don't know if that is the result of terminating the charge cycle at 3.95V, the discharge cycle at 2.96V or the pulse charging regime. When I get back on deck I will be able to finally start a proper trial to see if pulse charging is of a benefit.
More fully explained in this post but briefly I am using a control battery to confirm if any degradation is due to float charging or simply the passage of time and high ambient temperatures. Apart from the time being tested the test battery remains in a CC/CV charger floating at 3.93V. This charges the cell to approximately 70% of capacity then floats at this voltage. The control cell is charged to about 50% capacity before being removed from the charger and stored on the shelf. When the control battery is tested it is fully charged before the test.
Having established a base line for the cell capacity when fully charged I have now floated one cell at 3.93V for 12 months. During the test period the floating cell has been subject to 22 discharge tests from the float voltage to 3.0V, and a further 5 discharge tests when fully charged.
Looking at the discharge tests of the battery left on float charge at 3.93V there was no meaningful degradation in capacity.
A
full charge of the test battery versus the control battery shows no
ill-effects after 12 months. While one cell is not a rigorous sample it should make the self-proclaimed experts feel very foolish. But since they confuse trickle charging with float charging, which are two different charging regimes, and never substantiate any of their claims I suspect they are pretty much clueless all of the time.
If you know how to float charge a Li-Ion cell then there is no evidence that a
conservative float charge does any damage to the cell.
After 178 days the Anko cell was finally exhausted. That fell well short of the 410 days expected suggesting the nominal capacity was much less than 210mAh. While I only tested one cell from the pack of 4 I wouldn't at present recommend theses cells. I'm going to try one of these Panasonic cells next which I have next. These appear to have lasted very well (I use it several times a day it seems) in my implementation of this continuity tester.
I own 4 versions of the NanoNVA and have found it incredibly useful. However, I never stopped to question just how accurate or reliable the measurements are. That was until Matthew, VK5ZM, was in my workshop on the weekend and he made a suggestion in response to some concerns I was having over the measurement of attenuators. The outcome of following that suggestion was a better understanding of just how good the NanoVNA is and what might be limiting it's performance.
The assumption I implicitly made was the SOLT calibration process reduced errors in measurement to an immaterial level. Rather than lots of mathematics to validate or invalidate the NanoVNA you should repeat the following steps with your instrument to decide for yourself what you are prepared to accept as correct, close enough or simply wrong.
Ideally you will have a SMA attenuator, 10-15dB in value, that you can insert before the 50 ohm calibration load.
Perform a SOLT calibration directly on the SMA port, with the load being the attenuator / reference load combination and the Through step using one of the supplied cables. I used the NanoVNA app to average 8 readings and a 1 to 801MHz range with 401 or 801 steps.
Now reattach the attenuator / reference combination and measure the input return loss. Save that to perhaps M1.
Now remove the reference load leaving just the attenuator attached to the measurement port. You should see a chart like this:
The purple trace is the input return loss of the load plus attenuator combination and two things stand out: Noise ( even after averaging ) and the step change in measurements once the frequency was higher than 300MHz.
The red line is the attenuator alone and as expected the return loss was twice the attenuation (2x-10dB).
That red line confirms that input return loss measurements at the NanoVNA SMA connector down to -20dB can be accepted as correct, anything down to -45dB up to 300MHZ and perhaps -35db to -30dB above that is probably close enough. And any return loss measurements approaching the purple line, or below that, is probably invalid and you should be rounded up to whatever "close enough" reference you decide upon.
Save your calibration and then attach the attenuator / reference load combination at the point you normally calibrate to. For me that's at the end of a SMA to SMA cable some 25cm in length. It could be one of the supplied cables but I replaced mine since one became damaged.
Repeat the SOLT calibration at the end of the SMA cable (Reference plane) then measure the input return loss of the attenuator load combination, and the attenuator by itself.
My results are shown below:
The red line, attenuator only, is essentially unchanged but the attenuator / load combination had shifted upwards and noise has reduced.
At this point you should be starting to question why this is so. Is the cable lossy? Is this a consequence of the SOLT calibration? Why did this happen?
OK, it's not that bad but now you will see some results that should forever change your opinion on how good your measurements are.
Restore the calibration to the values saved when the attenuator / load were directly attached to the NanoVNA. Attach your SMA cable to the NanoVNA and the attenuator / load on the far end. Blink in disbelief and sit down. A 50 ohm load at the end of a 50 ohm cable should measure the same return loss as when the cable was not present. If the cable had loss then the return loss would be lower.
Blue SMA cable supplied with NanoVNA V4 in metal box (ie the supposedly top end version)
Clearly it's not a 50 ohm cable. Calibration means I never knew this cable was rubbish since calibration hides how bad this cable is. But it puts into question past measurements I made using this cable with the "Top End" NanoVNA model.
The black SMA to SMA cable supplied with the lower end NanoVNA is slightly better but still rubbish:
And now a collection of results as I tried to look for a good cable.
 |
| Long Cable SMA one end, N Connector the other |
 | ||
| Hardline - SMA to N |
 |
| Double shielded Silver plated - SMA to N (A quality cable) |
 |
| SMA to N Pigtail from AliExpress |
I now appreciate why high end equipment comes with very expensive
connecting cables. The intrinsic ability of the NanoVNA to measure input
return loss is quite good, especially below 300MHz. However, the cables
used can result in meaningless measurements below a certain threshold
despite the SOLT calibration. I seems I don't own a single cable with SMA connectors each end that I would accept for measurement purposes.
While the SOLT calibration can reduce the influence of cables and connectors up to the reference plane there will be a point where measurement uncertainty creeps. And if you flex, twist, curl or bend the cable between the reference plane and NanoVNA the cable impedance could, probably will, change which again introduces measurement uncertainty.
Measurement uncertainty means that when you are tuning / matching a load to 50 ohms is -35dB actually better than -30dB? Is a -20dB return loss the influence of cables/connectors or is it a true measurement of the load? I have noted from past testing with a homebrew return loss bridge that different examples of even the best quality 50 ohm cables I have terminated in N Connectors can show several dB of difference when measuring return losses.
If you want to reduce your measurement uncertainty you will need to:
Only then should you be comfortable relying on the measurements.
Whether you call it VSWR, R+jX, a smith chart or input return loss measuring reflections from the load is one of the key uses of the NanoVNA. It is well worth the few minutes it takes to test the cables so you understand when your measurements become uncertain.
I'm off to buy a -15dB SMA attenuator to improve my comfort level in the measurements being made. I'll post an update once I have procured one and repeated the tests.
I recently took ownership of some larger 20Ah LiFePO4 cells and the charging adventures that followed are worth sharing.
The first LiFePO4 cells I purchased that were actually delivered were 26700 sized with a claimed capacity of 4000mAh. Charging them for testing was easy enough with a variation of my CC/CV circuit. They never delivered 4000mAh.
Ditto the 14Ah cells I brought for my ride on mower. More like 12.5Ah in capacity they took more effort on my part but I have an old HP power supply on my bench that I can manually set up for CC/CV charging.
When I received some 20Ah cylindrical cells things started to go awry. I didn't like leaving my HP 20A power supply running unattended with a charging current greater than 2.5A because of the gauge wiring I was using to connect to the battery. However, I couldn't readily charge these cells at less than this. At lower currents the energy just seemed to disappear.
So I built heaver gauge charging cables and charged them while I was present. Surprisingly, the rated 20Ah capacity was achieved when discharged at 1.5A.
During all of this I looked for an alternative. I first tried some supposed LM2596 buck convertors. Advertised as 3A and 40V input I thought these would work really well with some 200W 30V solar panels I have. The first time I connected one to a 30V supply, with no load, for testing the module burnt out forcing a retreat outside till the fumes dissipated.
From a 12V supply they could deliver 2.5A to 3A but the heat they dissipated meant they could only be used intermittently. And some care is needed adjusting the trim-pot which is difficult to set within 20mV of 3.65V.
I had some server power supplies. One model, a HP unit, appeared suitable but the 3.3V rail could only be coaxed to 3.45V. I didn't persist with modifying the unit because I tried the 5V rail expecting it to shut down. Instead, it happily dumped +20A into the cell. Which gave me a different idea.
Take the 5V rail and then regulate it to 3.65V with some current control. Assume a 10A charging rate. 10A x (5-3.65)V means dissipating 13.5W of heat. That sounds achievable. Except the dropout voltage for a LM317 alone is 1.7V. So a purpose built regulator or a lot of time searching/testing would be needed. The design time for this approach seems high. It would be easy to do this from a 12V rail but the additional 70W of heat generated needs a big heat sink.
Then I found some LM1084 regulators. Similar to the LM317 but the drop out voltage is much lower (around 1V) and each one is good for 5A. I plucked a CPU heatsink from the junk box, mounted the regulator and tested the concept with heavy gauge wiring.
Partial success. I can now charge at a little over 4 amps continuously with no concerns over heat dissipation. My bench power supply has been released from battery charging duty.
At present I'm getting on with other projects. But my planned next steps are: