Ant switch pattern and IQ sampling

RE: Ant switch pattern and IQ sampling

Saleh — Thu, 29 Jul 2021 14:26:59 GMT

There is a comprehensive conversation available in this thread, just I try to write a fresh answer from my point of view, I hope it can help:

[quote userid="102154" url="~/f/nordic-q-a/76620/ant-switch-pattern-and-iq-sampling"]Why am i having 45 samples? I know there is a guard period and then as many samples as ant_pattern_length but it is 12 antenna long.[/quote]

Number of samples doesn't depend on the antenna pattern size, it is dependent on two parameters:

a. CTE duration

b. TSAMPLESPACING (Corresponds to the 1us or 2us sample slots)

The first parameter is chosen by the transmitter and can changed here, the default value is the maximum:

https://github.com/nrfconnect/sdk-nrf/blob/master/samples/bluetooth/direction_finding_connectionless_tx/src/main.c#L20

The second parameter is chosen by the receiver and can be modified here, the default value is 2us sample slot:

https://github.com/nrfconnect/sdk-nrf/blob/master/samples/bluetooth/direction_finding_connectionless_rx/src/main.c#L232

NOTE: At the time of writing this answer, there is a bug that prevents the receiver to work on 1us slot, an issue and fix has provided for this problem:

Issue: https://github.com/zephyrproject-rtos/zephyr/issues/37305

Fix: https://github.com/zephyrproject-rtos/zephyr/pull/37308

[quote userid="102154" url="~/f/nordic-q-a/76620/ant-switch-pattern-and-iq-sampling"]Since there are more value than i expect, should i make an algorithm in order to discard the wrong one ?[/quote]

The numbers of values is correct, you can calculate it as:

NumberOfValues = 8 + (CTELength - 12) / (2 * SampleSlot)

Hint: 12 is total length of "Gaurd Period" and "Refrence Period" and despite of the TSAMPLESPACING, it generates 8 sample per "Refrence Period".

e.g. for CTE with maximum length (160 us), and default value for sample slot (2 us), it will be:

8 + (160 - 12) / ( 2 * 2)

= 8 + 148 / 4 = 8 + 37 = 45

[quote userid="102154" url="~/f/nordic-q-a/76620/ant-switch-pattern-and-iq-sampling"]s the values correct ? i know the swing of IQ value are bigger than those.[/quote]

At least I am receiving the same values, but I am not sure about it and should be evaluated with a reference, later.

[quote userid="102154" url="~/f/nordic-q-a/76620/ant-switch-pattern-and-iq-sampling"]Last but not least, where can i find in example project all the parameter such as TSAMPLESPACINGREF / TSAMPLESPACING / TSWITCHSPACING ?[/quote]

These parameters are set by lower layers and is not accessible in application layer. Also it is not recommended to modify these values in lower layers, because they are set by experienced developers for each chip-set.

but instead you can change the sample slot duration at this line as I mentioned before, it consequently will change those parameters in the lower layers:

https://github.com/nrfconnect/sdk-nrf/blob/master/samples/bluetooth/direction_finding_connectionless_rx/src/main.c#L232

Regards,

Saleh

RE: Ant switch pattern and IQ sampling

Elfving — Mon, 05 Jul 2021 13:10:38 GMT

Hi Nathan!

Sorry for not getting back to you sooner, I've been away for a week. Once again I will try to answer your questions, but our support for this is very limited.

What you've found in the Bluetooth Spec and deduced from it seem correct. I think the reason it loops with an interval of 11 instead of 12 is that the first antenna is used in the reference period, and the looping pattern reverts to the one that started the sample period. This is what they define as "the beginning" in the spec. See figure 12, section 4 in the whitepaper.

I'm glad to hear that you fixed the sampling issue. I'll have to get back to you on saturation and gain of IQ samples.

Best regards,

Håkon

RE: Ant switch pattern and IQ sampling

Nathan_INOSI — Mon, 28 Jun 2021 10:21:33 GMT

Hi Ekfving,

I found a bug which block switching to slot_durations 0x1.

The setup : direction finding locator and beacon, aoa mode only, change slot_durations from 0x2 to 0x1

The output from locator :

Creating Periodic Advertising Sync...success.
Waiting for periodic sync...
PER_ADV_SYNC[0]: [DEVICE]: 21:9D:41:2C:45:8D (random) synced, Interval 0x0780 (2400 ms), PHY LE 2M
success. Periodic sync established.
Enable receiving of CTE...
failed (err -22)
Scan disable...Success.
Waiting for periodic sync lost...
PER_ADV_SYNC[0]: [DEVICE]: 21:9D:41:2C:45:8D (random), tx_power 127, RSSI -44, CTE AOA, data length 0, data: 
PER_ADV_SYNC[0]: [DEVICE]: 21:9D:41:2C:45:8D (random), tx_power 127, RSSI -44, CTE AOA, data length 0, data:

The failed error prevent IQ sampling. The problem come from function validate_cte_rx_params in direction.c. If i put in comment the section about slot_durations check (222-226), the failed error disappear.

static int validate_cte_rx_params(const struct bt_df_per_adv_sync_cte_rx_param *params)
{
	if (!(params->cte_type &
	      (BT_DF_CTE_TYPE_AOA | BT_DF_CTE_TYPE_AOD_1US | BT_DF_CTE_TYPE_AOD_2US))) {
		return -EINVAL;
	}

	if (params->cte_type & BT_DF_CTE_TYPE_AOA) {
		if (df_ant_info.num_ant < DF_SAMPLING_ANTENNA_NUMBER_MIN ||
		    !BT_FEAT_LE_ANT_SWITCH_RX_AOA(bt_dev.le.features)) {
			return -EINVAL;
		}

		/*if (!(params->slot_durations == BT_HCI_LE_ANTENNA_SWITCHING_SLOT_2US ||
		      (params->slot_durations == BT_HCI_LE_ANTENNA_SWITCHING_SLOT_1US &&
		       DF_AOA_RX_1US_SUPPORT(df_ant_info.switch_sample_rates)))) {
			return -EINVAL;
		}*/

        // Workaround
        if (!(params->slot_durations == BT_HCI_LE_ANTENNA_SWITCHING_SLOT_2US ||
		      params->slot_durations == BT_HCI_LE_ANTENNA_SWITCHING_SLOT_1US)) {
			return -EINVAL;
		}

		if (params->num_ant_ids < BT_HCI_LE_SWITCH_PATTERN_LEN_MIN ||
		    params->num_ant_ids > df_ant_info.max_switch_pattern_len || !params->ant_ids) {
			return -EINVAL;
		}
	}

	return 0;
}

It is the "params->slot_durations == BT_HCI_LE_ANTENNA_SWITCHING_SLOT_1US && DF_AOA_RX_1US_SUPPORT(df_ant_info.switch_sample_rates)" that don't return 1, and therefore return error -EINVAL.

In my build directory, in .conf, the CONFIG_BT_CTLR_DF_ANT_SWITCH_RX and CONFIG_BT_CTLR_DF_CTE_RX_SAMPLE_1US are set.

Regards,

Nathan

RE: Ant switch pattern and IQ sampling

Nathan_INOSI — Fri, 25 Jun 2021 13:40:35 GMT

Hi Elfving,

Thanks for the answer. I learn a lot with those info. I have now some explanation about the parameters, correct me if i'm wrong.

In Bluetooth Core v5.2, 6/B/2.5.1 - "The Constant Tone Extension has a variable length; it shall be at least 16 μs and not greater than 160 μs." That explain the choice of 160µs (0x14U in tx exemple) for the CTE length.

In Bluetooth Core v5.2, 6/B/2.5.4 - "the receiver shall take an IQ sample each microsecond during the reference period and an IQ sample each sample slot (thus there will be 8 reference IQ samples, 1 to 37 IQ samples with 2 μs slots, and 2 to 74 IQ samples with 1 μs slots, meaning 9 to 82 samples in total)." Since the rx example is configured at 2µs by default, we are in the 8+37 samples case which make 45 samples. So as you say, i do not need to discard any sample since they are all usefull.

In Bluetooth Core v5.2, 6/A/5.1 - "The first antenna in the pattern shall be used during the reference period [...]. The second antenna in the pattern shall be used during the first sample slot, the third antenna during the second sample slot, and so on. The same antenna ID may appear more than once in the pattern. [...] If the pattern specified by the Host is exhausted before the last sample slot, it shall be restarted from the beginning..." Since i use the 12 antenna patern from Nordic, i expect the pattern to be restart 3 times during the 45 samples. Even with saturation, we can see it in my plot from IQ sampling. But, for some reason, the antenna pattern seem to be 11 antenna long because between two spike i've got 11 IQ samples.

In zephyr/subsys/bluetooth/controller/ll_sw/nordic/hal/nrf5/radio/radio_df.c

void radio_df_ant_switch_pattern_set(uint8_t *patterns, uint8_t len)
{
	/* SWITCHPATTERN is like a moving pointer to underlying buffer.
	 * Each write stores a value and moves the pointer to new free position.
	 * When read it returns number of stored elements since last write to
	 * CLEARPATTERN. There is no need to use subscript operator.
	 *
	 * Some storage entries in the buffer has special purpose for DFE
	 * extension in radio:
	 * - SWITCHPATTERN[0] for idle period (PDU Tx/Rx),
	 * - SWITCHPATTERN[1] for guard and reference period,
	 * - SWITCHPATTERN[2] and following for switch-sampling slots.
	 * Due to that in SWITCHPATTER[0] there is stored a pattern provided by
	 * DTS property dfe_pdu_antenna. This limits number of supported antenna
	 * switch patterns by one.
	 */
	NRF_RADIO->SWITCHPATTERN = DT_PROP(RADIO, dfe_pdu_antenna);
	for (uint8_t idx = 0; idx < len; ++idx) {
		NRF_RADIO->SWITCHPATTERN = patterns[idx];
	}
}

SWITCHPATTERN[0] = ANT 12 (idle period)

SWITCHPATTERN[1] = ANT 11 (guard and reference period)

SWITCHPATTERN[2..11] = ANT 12 - 1 - 2 - 10 - 3 - 9 - 4 - 8 - 7 - 6 - 5

Could i be that the actual switch pattern repeating is patterns[2...11] and could explain my 11 antenna pattern ?

Two more thing to add :

i didn't find any IQ sample info about satuation. Do you have any link ?
If we change slot_durations to 1, the receiver don't sample and print nothing.

Regards

Nathan

RE: Ant switch pattern and IQ sampling

Elfving — Thu, 24 Jun 2021 14:10:57 GMT

The spiking antennas are probably caused by multipath issues, this needs to be dealt with. The gain might be the main culprit, that needs to be modified such that the signal doesn't go to saturation. If it goes to saturation, the entire series should be discarded. There should be a field in the IQ samples that show saturation.

Regarding your previous questions:

1. This depends on the sample_count, which depends on sample rate and the length of the CTE. The CTE length can be found and edited in the beacon example, seems it is 160us by default. As you can see in section 4 of the whitepaper, the algorithm basically continues at the start of its antenna switching pattern if it runs out of antennas.

2. Nothing needs to be discarded per se.

3. They seem ok, besides what I mentioned above about spikes.

4. The parameters can be found by using ctrl+shift+F in SES, maybe most notably in zephyr/subsys/bluetooth/controller/ll_sw/nordic/hal/nrf5/radio/radio_df.c.

Best regards,

Elfving

RE: Ant switch pattern and IQ sampling

Nathan_INOSI — Wed, 23 Jun 2021 13:39:15 GMT

Hi Elfving,

Where can i find the sample_count variable ?

I've made progress in my understanding. I know that :

the amplitude = sqrt(pow(I,2)+pow(Q,2))
the phase = arctan2(Q,I)

So i decide to plot the amplitude in order to see any incoherence. (sorry for the bad resolution)

I put the beacon close to the locator when i reccord the sample. I know for sure that the 8 first sample are the reference, then it's switch slot and then sample slot. I guess the value close to the reference value are valid and all other sample are taken when the module switch antenna.

I can't find where the code setup the sampling, can you help me ?

Thanks

Nathan

RE: Ant switch pattern and IQ sampling

Elfving — Wed, 23 Jun 2021 12:53:30 GMT

Hi Nathan!

Sadly our support on direction finding is quite limited, but I'll get back to you on this as soon as I can find answers for you.

From what i can see right now though, the IQ values seem fine. And the number of samples depends on the sample_count variable, not the number of antennas.

Best regards,

Elfving

RE: Ant switch pattern and IQ sampling

Zack_Lee — Wed, 23 Jun 2021 06:45:44 GMT

Hi Nathan,

may i know how to add overlay-aoa.conf to CMAKElist as you mention above? are you adding both to RX and TX or just TX will do?