In LTE, usually they use multiple Antenna for downlink (at least from Category 2 UE and higher), meaning that eNode (Network) has use multiple Tx Antenna and UE use multiple Rx antenna.

Now you almost automatically think about 'MIMO', but in reality 'multiple antenna' does not automatically mean 'MIMO'. For example, you have two downlink antenna. You can use these two antenna in various ways. Of course, one ways is to use it as 2 x 2 MIMO, but this is not the only way. You can use the two antenna in diversity configuration rather than MIMO configuration. Or you can just use only one of the antenna and sometimes you would like to use various different multiplexing, precoding methods etc.

In LTE, they give a special name for each of the way of transmission and it is called 'Transmission Mode'. For example, what we normally call 'SISO' (Single Transmission Antenna and Single Reciever Antenna) is called 'TM1(Transmission Mode 1)'. What we normally call 'Diversity' is called 'TM2'.  What we call 'MIMO' but no PMI feedback from UE is called 'TM3'. MIMO and UE feedback from UE (CQI, PMI, RI) is called 'TM4'.

NOTE : Regarding how each of various transmission mode is utilized in different situations and how they are configured in signaling message, check out this tutorial of Amarisoft TechAcademy.

Let's follow the TMs from their definition in 36.213 down to the physical layer path, and then look at measured BLER and a drive test log. Keep one thing in mind throughout. A TM is configured per UE by RRC, but the DCI format in each subframe still decides which of the allowed PDSCH schemes is used.

The page covers the following topics.

• Which PDSCH scheme does each TM assume for CQI ?
• Which DCI format and PDSCH scheme go with each TM ?
• How does a TM choose the path through the physical layer ?
• Transmission Mode and BLER
• Example 1 - Transmission Mode changes while driving in live network
• Transmission Mode and Reference Signal - Antenna Ports
• References

Which PDSCH scheme does each TM assume for CQI ?

A UE reports CQI for a hypothetical PDSCH transmission, so it needs to know which transmission scheme to assume. 36.213 Table 7.2.3-0 fixes that scheme for each TM. Reading it row by row is also the shortest summary of what each TM does.

A good summary of each Transmission Mode can be as following table from 36.213. These table ecolved and got more complicated as LTE evolves. I have listed tables from different releases for comparison.

< 36.213-Table 7.2.3-0 : PDSCH transmission scheme assumed for CQI reference resource - Rel 8 >

< 36.213-Table 7.2.3-0 : PDSCH transmission scheme assumed for CQI reference resource - Rel 10 >

< 36.213-Table 7.2.3-0 : PDSCH transmission scheme assumed for CQI reference resource - Rel 12 >

< 36.213-Table 7.2.3-0 : PDSCH transmission scheme assumed for CQI reference resource - Rel 13 >

• Rel 8 : TM1 to TM7. TM1 is single-antenna port 0 and TM2 is transmit diversity. TM3 switches between transmit diversity and large delay CDD by rank. TM4 is closed-loop spatial multiplexing, TM5 is multi-user MIMO and TM6 is closed-loop spatial multiplexing with one layer. TM7 assumes port 0 or transmit diversity, depending on the number of PBCH antenna ports.
• Rel 10 : the green box marks TM8 and TM9. Both depend on whether the UE is configured with PMI/RI reporting. With PMI/RI reporting, TM9 assumes up to 8 layer transmission on ports 7-14.
• Rel 12 : the green box marks TM10. TM10 follows the same logic as TM9, but per CSI process and based on the number of CSI-RS ports.
• Rel 13 : the green box marks the TM9 changes. The text is split into non-BL/CE and BL/CE cases, and for a BL/CE UE the assumption depends on periodic CSI reporting mode 1-1.

The table did not stop at Rel 13. In 36.213 v19.4.0 its title reads "PDSCH transmission scheme assumed for CSI reference resource", and the TM9 and TM10 rows now begin with "For activated serving cells". Both rows also add a case for dormant serving cells. For a dormant serving cell, the UE assumes single-antenna port 0 if the number of PBCH antenna ports is one, and otherwise transmit diversity.

Notice the pattern across releases. TM1 to TM7 never changed. Each new TM added a condition on the CSI configuration, because TM8, TM9 and TM10 rely on UE-specific reference signals and the CQI assumption has to follow what the UE is configured to report.

• Table 7.2.3-0 is a one-line summary of each TM : it gives the PDSCH scheme the UE assumes when it calculates CQI.
• TM8 to TM10 depend on the CSI configuration : PMI/RI reporting and the number of CSI-RS ports decide the assumed scheme.
• The latest version has a dormant SCell case : for a dormant serving cell, TM9 and TM10 fall back to port 0 or transmit diversity.

Which DCI format and PDSCH scheme go with each TM ?

The TM alone does not tell the UE how a PDSCH is sent in a given subframe. The DCI format does, and each TM allows exactly two formats with C-RNTI: the fallback DCI format 1A and one format specific to the TM. 36.213 Table 7.1-5 lists those pairs.

Considering these various possibilities, 3GPP provides several predefined transmission methods and this transmission method is called 'Transmission Mode'. For now, there are seven predefined predefined transmission mode as shown in the following table (TS 36.213)

That count of seven is from Rel 8. TM8 came in Rel 9, TM9 in Rel 10 and TM10 in Rel 11, as the suffixes tm8-v920, tm9-v1020 and tm10-v1130 in 36.331 show. The Rel 10 and later tables below therefore carry more rows than the Rel 8 table.

< 36.213 - Table 7.1-5 : PDCCH and PDSCH configured by C-RNTI - Rel 8>

< 36.213 - Table 7.1-5 : PDCCH and PDSCH configured by C-RNTI - Rel 10>

< 36.213 - Table 7.1-5 : PDCCH and PDSCH configured by C-RNTI - Rel 12>

< 36.213 - Table 7.1-5 : PDCCH and PDSCH configured by C-RNTI - Rel 13>

< 36.213 - Table 7.1-5A: EPDCCH and PDSCH configured by C-RNTI - Rel 13>

With LTE M1 (BL/CE) in Rel 13,  a new table was added as follows.

< 36.213 - Table 7.1-5B: MPDCCH and PDSCH configured by C-RNTI - Rel 13>

• Table 7.1-5 : DCI format 1A is allowed in the common and the UE specific search space for every TM. The TM-specific format is allowed only in the UE specific search space.
• Rel 10 and Rel 12 tables : TM9 adds DCI format 2C and TM10 adds DCI format 2D, both for up to 8 layer transmission on ports 7-14. In an MBSFN subframe, DCI format 1A in TM9 and TM10 means single-antenna port 7.
• Rel 13 Table 7.1-5 : DCI formats 2C and 2D also allow single-antenna port 7, 8, 11 or 13 when the UE is configured with dmrs-tableAlt.
• Table 7.1-5A : the same TM and DCI pairs apply when the PDSCH is scheduled by EPDCCH, but both formats sit in the UE specific search space.
• Table 7.1-5B : a BL/CE UE uses only TM1, TM2, TM6 and TM9, scheduled by MPDCCH with DCI formats 6-1A and 6-1B.

The table below puts the TM-specific DCI format and the release of each TM side by side. The release comes from the ENUMERATED values of transmissionMode in AntennaInfoDedicated and AntennaInfoDedicated-r10 in 36.331 v19.3.0.

36.213 v19.4.0 still defines mode 1 to mode 10, so no TM has been added since Rel 11. It does add Table 7.1-5C for short TTI, where SPDCCH schedules PDSCH with DCI formats 7-1A to 7-1G. For frame structure type 1, the short TTI TMs are 1, 2, 3, 4, 6, 9 and 10 in non-MBSFN subframes, and frame structure type 2 adds TM8.

• Every TM pairs DCI 1A with one TM-specific format : DCI 1A is the fallback, usually transmit diversity or port 0.
• TM8, TM9 and TM10 came after Rel 8 : 36.331 marks them tm8-v920, tm9-v1020 and tm10-v1130.
• BL/CE UEs use a reduced set : only TM1, TM2, TM6 and TM9 with MPDCCH and DCI formats 6-1A and 6-1B.

How does a TM choose the path through the physical layer ?

A TM is, in the end, a choice of path through three blocks of the PDSCH processing chain. Let's see which blocks those are, and then check the codewords, layers and codebook entries each TM uses.

To understand very details of each transmission mode requires almost complete knowledge of physical layer processing. Three important blocks in physical layer to determin the transmission mode can be illustrated as follows. You will find many different ways from the data input (left most arrow) through the final antenna ports (rightmost arrows). Each transmission mode determin which path the input data should follow through.

• Channel Coding : the input data becomes Codeword 0, or Codeword 0 and Codeword 1.
• Layer Mapper : the codewords are mapped onto Layer 0 and Layer 1. The red dashed lines show that one codeword can go to one or both layers.
• Precoding : the layers are mapped onto Antenna port 0 and Antenna port 1, again with every combination drawn as a dashed line.

Some important parameter sets for each transmission mode are as follows. (To understad this process in detail, it is crucial to understand details of Precoding in basic procedure page).

Read the table with one rule from 36.211 in mind. For transmit diversity, the number of layers equals the number of antenna ports, so the transmit diversity rows show 2 layers for 2 ports. For closed-loop precoding without CDD, the first number in {Number of layers, Codebook index} is the layer count. So the rank-1 rows of TM4, TM5 and TM6 carry one layer, and TM4 uses two layers only with two codewords.

TM7 and TM8 are different. Their spatial multiplexing and single-port rows do not use a codebook from Table 6.3.4.2.3-1, because the UE demodulates with UE-specific reference signals. The eNB can apply any precoding it wants, and the UE sees the result as a single port or as two ports.

• A TM fixes the path from codewords to antenna ports : codeword count, layer count and precoding are chosen together.
• Transmit diversity uses as many layers as antenna ports : that is why TM2 shows 1 codeword on 2 layers.
• Codebook-based TMs show the rank in the codebook entry : {1, x} is one layer and {2, x} is two layers.
• TM7 and TM8 need no codebook : the UE-specific reference signal carries the precoding with the data.

Transmission Mode and BLER

Each TM trades throughput against robustness in a different way, and BLER against SNR shows the robustness side. Treat the four plots in this section as trends for one set of assumptions, not as numbers for a real network.

There are various purpose (reasons) for each different Transmission Mode. Some TM is designed mainly to increase throughput, some are to increase communication reliability and some are to handle multiple users simultaneously etc. In this section, I would show you some examples showing the increased communication reliability with a couple of different Transmission Mode. Followings are four plots from Ref [4].  

First compare (A) with (B). In terms of throughput, they (TM1 and TM2) are same. But you see much higher performance in terms of BLER measurement in TM2. This would match the common sense that you have. The major purpose of Diversity is to achieve more robust data decoding and decrease error rate (If you are not familiar with interpreting this kind of graph, refer to SNR page. In terms of interpreting the general meaning of the graph, you will get some insight from the interpretation of SNR vs BER graph).

Then, Compare (A) and (C), i.e, compare TM1 (SISO) and TM3(MIMO). The major purpose of MIMO is to increase the throughput (but throughput part is not shown in this graph). However, you would see much increased error rate in (C). Note : Be careful, the range of horizontal axis of (A) is different from other three graphs. So it may be a little confusing.

Then compare (C) and (D). In this case, Transmission Mode are same, but the antenna configuration is different. (C) is 2x2 MIMO and (D) is 4x2 MIMO. At very high level view, 4x2 is a kind of diversity for 2x2. Accordingly, you will see lower error rate (Diversity gain) in (D) comparing to (C).

• (A) TM 1, 1 Tx and 1 Rx : BLER reaches 10^-1 at about 12 to 14 dB SNR and is still about 2x10^-2 at 20 dB. The horizontal axis runs from -10 to 30 dB.
• (B) TM 2, 2 Tx and 2 Rx : BLER falls to 10^-1 at about 4 to 5 dB and to 10^-3 at about 11 dB.
• (C) TM 3, 2 Tx and 2 Rx : the curve starts at 10 dB and reaches 10^-2 at about 19 dB.
• (D) TM 3, 4 Tx and 2 Rx : the curve starts near 9 dB and reaches 10^-2 at about 14 dB, several dB earlier than (C).

Put (A) and (B) next to each other and the diversity gain is large. TM2 reaches 10^-1 roughly 8 to 10 dB earlier than TM1, because the two copies of the signal rarely fade at the same time. TM3 in (C) needs much more SNR for the same BLER, because it sends two layers and each layer gets only part of the power and the diversity. The extra two transmit antennas in (D) restore part of that diversity.

• TM2 improves robustness, not rate : the same data is sent over two antennas, so BLER drops at a given SNR.
• TM3 improves rate, not robustness : two layers need higher SNR for the same BLER than one layer.
• More transmit antennas than layers adds diversity : 4x2 TM3 reaches the same BLER at lower SNR than 2x2 TM3.

Example 1 - Transmission Mode changes while driving in live network

In a live network the TM is not fixed for the whole session. The eNB can reconfigure it with RRC as the UE moves, and a drive test log shows how often that happens.

When a UE drive through multiple different networks (e.g, during driving test), it may experience various transmission mode. Let's take a look at a measurement done by a UE in the field. Following plot is from the data captured by a drive test tool Azenqos Drive Test tool (AZQ Android). I got the log captured by the tool and exported the data as csv file and then plot it on Microsoft Excel and figured out the transmission mode setting based on RRC Connection Configuration message that are captured by the same tool.  It is almost impossible to correlate the transmission mode and throughput performance just from the following plot because there are so many factors are involved. My intention is to show you how dynamically the transmission mode changes in the live network.

• Dashed lines : each vertical line marks the point where the TM changed. Black is TM2, blue is TM3 with p-a dB-3 and red is TM4 with p-a dB0.
• Top panel : LTE L1 Througput Mbps arg(1), mostly below 40 Mbps, with a peak above 60 Mbps.
• Middle panel : LTE BLER arg(1), mostly between 5 and 15 percent.
• Map : the drive route, with the same TM change points drawn across it.

Notice that p-a changes together with the TM. p-a in PDSCH-ConfigDedicated sets the PDSCH EPRE relative to the cell-specific RS EPRE. So a TM change in this log also changes the PDSCH power, and a throughput change at the same point cannot be attributed to the TM alone.

TM2 appears only at the start of the log. After that, the network switches between TM3 and TM4 many times. The BLER plot shows no clear step at the change points, which agrees with the author's note that the plot alone cannot link a TM to its throughput.

• The TM changes often in a live network : the network reconfigured it many times along a short drive.
• A TM change can come with a power change : p-a is dB-3 with TM3 and dB0 with TM4 in this log.
• Throughput differences need controlled tests : a drive test mixes the TM with radio conditions and load.

Transmission Mode and Reference Signal - Antenna Ports

A TM also decides which reference signal the UE uses to demodulate the PDSCH. That link is often the source of confusion, because the antenna port numbers change between TMs.

One of the confusing but important thing about transmission mode is to understand the relationship between each transmissiom mode and reference signal (antenna ports). For this, refer to Reference Signal (Antenna Port Number) vs Transmission Mode in Reference Signal (Downlink) page.

In short, TM1 to TM6 demodulate the PDSCH with the cell-specific reference signal on ports 0 to 3. With the TM-specific DCI format, TM7 uses the UE-specific reference signal on port 5, and TM8 uses ports 7 and 8. TM9 and TM10 use ports 7 to 14 for demodulation and rely on CSI-RS for channel state information. The DCI tables above show the same port numbers in their PDSCH scheme column.

This is also why the codebook column of the parameter table is empty for TM7 and for the UE-specific RS rows of TM8. The UE estimates the precoded channel directly from the UE-specific reference signal, so it does not need to know which precoder the eNB used.

TM10 adds one more step on top of TM9. A UE in TM10 can be configured with up to 8 parameter sets, and the PDSCH RE Mapping and Quasi-Co-Location Indicator field in DCI format 2D selects one of them. The parameter set tells the UE how the PDSCH is mapped around other reference signals and which CSI-RS the UE-specific RS is quasi co-located with. So the PDSCH can come from a different transmission point without any change in the TM.

• TM1 to TM6 demodulate with CRS : the CRS ports 0 to 3 carry the channel estimate.
• TM7 to TM10 demodulate with UE-specific RS : port 5 for TM7, ports 7 and 8 for TM8, and ports 7 to 14 for TM9 and TM10.
• TM9 and TM10 use CSI-RS for CSI : channel state is measured on CSI-RS, not on CRS.
• TM10 ties the PDSCH to a configured parameter set : DCI format 2D selects the RE mapping and quasi co-location for each PDSCH.

References

Here goes some additional material to read from LteUniversity. You would get pretty good high level picture from them.

[1] Transmission Mode #1 Or is it?

[2] Why Should You Care about Transmission Modes?

[3] Transmission Mode 3 The first of the 4 MIMOs!

[4] Downlink SNR to CQI Mapping for Different Multiple Antenna Techniques in LTE by Mohammad T. Kawser et al.

[5] TS 36.213 v19.4.0 : E-UTRA Physical layer procedures

[6] TS 36.211 v19.3.0 : E-UTRA Physical channels and modulation

[7] TS 36.331 v19.3.0 : E-UTRA Radio Resource Control (RRC) protocol specification

[8] TS 36.306 v19.3.0 : E-UTRA UE radio access capabilities