Pro Tools features three different audio engines and each one of them behaves differently. To illustrate this let's assume that I am buying an old HD Accel system because it is financially interesting. The HD Accel TDM mixer comes with a few more considerations and drawbacks such as having to be very careful on your gain stage to maintain healthy levels as 24-bit fixed-point TDM plugins only have a dynamic range of 144 dB and can therefore discard information or clip more easily. In the next sections I will discuss some good practices and optimizations to make the most out of Pro Tools mixers. AAX-DSP, AAX Native, and RTAS use a 32-bit float instead of a fixed 24-bit integer.
First, let's have a look at the different technologies involved with Pro Tools and a bit about plugin architecture. In modern digital audio, audio is sampled at 24 bits integer per sample by the audio interface' converters, and since each bit represents 6 decibels of dynamic range, the formula is:
Dynamic Range (dB) = number of bits ^ 6
24 bits ^ 6 = 144 dB
Since each bit can have two states, 0 or 1, the maximum resolution of a system is based on the number of values that can be represented for each sample. In an integer environment it is given by:
Resolution = 2^n (where n is the total number of bits)
2^24 = 16,777,216 values for a 24-bit system
So, in a 24-bit fixed point integer system, any number can be represented by one of the 16,777,216 values. One important limitation lies in the finite resolution of the representation. To illustrate this let's take the example of a 4-bit system, which gives 16 possible values (2^4). I can therefore represent number 0 to 15 in steps of 1. If I would like to represent 2.7, I cannot; I would have to round the result to the closest integer, adding distortion to the signal. After audio is sampled at 24-bit integer by the audio-to-digital converter (A/D), it is represented inside Pro Tools as 32-bit floating point for greater dynamic range representation and processing precision.
A 32-bit float system uses a 32-bit word made of a 24-bit mantissa to which we added an 8-bit exponent. While still using a 24-bit word, it has the ability to shift the exponent to represent the best possible value within the full 24-bit scale constraint, increasing the resolution and dynamic range of the system. While rounding errors also occur with floating point calculations, they are a lot smaller due to the increase in representation precision, especially for a 64-bit float.
Note
As an example to better understand floating-point mathematics, we can represent a number in two different ways: 1,000 or 1 x 10^3. Here, "1" is called the mantissa and "10^3" the exponent. The sign can be stored using different methods that we will not discuss in this book.
With integer "fixed point" calculations, if the number of bits is exceeded, we either clip or lose low-level information. A floating point system can "float" its decimal to achieve optimum signal representation, shifting the 24-bit word up, losing low-level information if needed but avoiding clipping, or like in the following diagram, shifting it down to increase the dynamic range of the signal. So if a number falls outside the fixed range, it can be "floated" by having its decimal shifted to make use of all available bits of the 24-bit word.
The IEEE 754 standard defines the specifications for single-precision (32-bit float) and double-precision (64-bit float) computations. The dynamic range of floating point systems is very hard to quantify. The mathematics can have some surprising behaviors, and floating precision tends to lose precision in some cases. At first, many users were unsure whether integer or floating-point calculations are the best for audio-related tasks, even though the theoretical benefits are beyond any doubt. Both representations have advantages and disadvantages that we will not discuss further in this chapter because they are not relevant. However, I would like to state the generally accepted approximate dynamic range value of 1,500 dB for single-precision (32-bit float) and 12,300 dB for double-precision (64-bit float). Precision computation is a direct reflection of the mantissa's word length and is therefore greatly increased by 64-bit float processing.
Now that we understand better the fundamental differences between integer fixed point and floating point representations, the following diagram will illustrate how the TDM plugins interact with the Pro Tools mixer against their AAX-DSP, AAX, and RTAS counterparts.
The preceding diagram represents the communication between the mixer and plugins. As shown, the left-hand side represents TDM, while the right-hand side of the diagram represents AAX-DSP/AAX/RTAS. When communicating with the internal mixer, information must be converted to floating point format. When coming from TDM fixed point calculation, information can be lost during the process; other formats will always retain full 24-bit of dynamic range.
When using AAX 32-bit, AAX-DSP 32-bit, or RTAS plugins, the 16- or 24-bit audio file is represented using a 32 bit float resolution at every sample, processed internally at higher resolution or not. It is then returned to the 32-bit float insert and summed with other signals inside the 64-bit float mixer using a 32-bit float resolution inserts on the master output tracks. We now begin to understand that there is a lot of conversion going on there, effectively "coloring" our signal.
If you are using HDX or HD Accel, it gets a bit more complicated. With AAX-DSP, same as native, everything stays at 32-bit floats, but with TDM, we use a 24-bit fixed integer. The signal can still be treated using higher resolution internally, but the plugin output will be returned to 24-bit to be mixed inside the 48-bit fixed point mixer. We can also use RTAS on an Accel system so more conversion steps can be involved when chaining different plugin types one after the other.
Floating point representation gives you enough headroom not to worry about clipping, but with a 24-bit fixed, anything below -144 dBfs will be lost and anything above 0 dBfs will be clipped. It becomes really important to be fully aware of those TDM limitations. The same applies for RTAS plugins used within an HD Accel environment since the 48-bit mixer communicates with plugins on a 24-bit bus. If you are using HD Accel, here are two good ways to do it:
Use master faders and watch the track meter to adjust bus levels before entering any plugin on the bus or on the audio track.
Look at the plugin clip indicator by looking at your channel strip. If any plugin' name is displayed in red, it has been clipped.
If using any other system, clip indicators inside the plugin might light up, but if the plugin is internally coded in 32-bit floats or higher, you can stack as many plugins as you like without ever needing to clip the audio, even if the internal meter shows clipping. On the other hand, even if you hear no distortion, it is also good practice to control your gain stage so you can also use these as mixing references.
Pro Tools 11 goes further, upgrading processing to 64-bit, which according to IEEE standards, is composed of one sign bit, a 52-bit mantissa, and an 11-bit exponent. That means even greater precision across the mixer and greater headroom than 32-bit floats. A 64-bit float precision increases fidelity across the entire signal chain, eliminating the need for signal representation changes. It has greater mantissa precision and enough dynamic range for you to never hear any audible distortion, and the signal will be very precise. Only the master output faders can show audible clipping as they represent the signal being sent to the digital-to-audio converters (D/A), which are 24-bit integers.
A 32-bit float was great but Pro Tools 11 takes audio fidelity to the next level by increasing the dynamic range and precision even further as well as by taking away all the unnecessary signal conversion along the entire signal path.
Voices are audio connection points between your CPU and the DSP, they represent audio tracks. Time slots or "Timeslots" mean two different things:
For HD Accel they represent connections not only within the DSP but also from the CPU or to the physical outputs of your interface, much like a patch bay. Timeslot numbers are fixed to 512 to move audio within the Pro Tools mixer, including TDM plugins. We can appreciate how fast their number can grow and possibly limit the system.
For HDX, they represent connection between multiple HDX cards meaning that an HDX1 system do not require any timeslots. This is because HDX uses new FPGA technology, moving away from full TDM architecture. The improved 1,536 timeslots TDM bus is only used to communicate between cards.
Pro Tools HD Accel has a maximum of 192 voices; HDX increases this number to 768. Under normal use, a mono track is equal to one voice, which means as long as you stay within the DSP domain. Once an audio stream is transferred into the Pro Tools mixer, it can leave the DSP via the interface or the track inserts for RTAS or AAX native processing. The latest option requires extra voices because we are moving back and forth from DSP, so we should be careful not to insert native plugins after DSP ones. As long as you respect this order, maximum efficiency can be achieved. Inside the playback engine setup, we can choose between different voicing settings. There will always be a combination of amount of voices to play back the tracks against available DSP usage for plugin processing.
Note
Time slots and voice counts are related to the sample rate. A higher sample higher than 48 kHz will bring down the number of voices and time slots available.
Here are some insert signal flow examples for a mono signal, to show how different arrangements can bring extra voice usage:
RTAS/AAX → AAX-DSP/TDM → Output = 1 voice
RTAS/AAX → AAX-DSP/TDM → RTAS/AAX → Output = 3 voices
Hardware inserts take two voices
An auxiliary input to a physical output takes one voice
An auxiliary return from a physical output takes one voice
When it comes to keeping everything in the box, we are always fighting with latency. Even mixing can be unresponsive when using MIDI controllers with large buffer sizes. The normal production process usually begins with recording, followed by adding plugins, and mixing later on. Sometimes recording an overdub within a busy mix session is required and can have different strategies depending on which Pro Tools setup we are using.
When working with accelerated systems such as HDX or Accel, we do not have to worry about setting up anything as all routing and audio processing is handled by the DSP; the latency is kept low when used with DSP plugins only. This is because only native plugins have to use the playback engine buffer; AAX-DSP or TDM plugins run natively on the HDX or Accel cards at low latencies all the time.
With all native solutions, the CPU handles the mixer and processing, so everything goes through the CPU H/W buffer, and therefore, any armed track will be affected by an increase in monitoring latency. Pro Tools provides a dedicated low latency monitoring option that can be configured from two different locations depending on what hardware you are using. For all systems, the feature can be enabled or disabled from the Option menu. For Avid-accelerated systems it is also configurable from Setup | I/O | Outputs.
Note
LLMs do not work if the track output is assigned to an internal bus. It has to be routed directly to the default 1 and 2 or user defined outputs.
Any native plugin and active auxiliary send associated with the track will be disabled.
Different native solutions offer different options as follows:
When using an HD Native card with Pro Tools HD 10 with a high H/W buffer size, we have to use the low latency monitoring mode. This option allows the user to choose a desired output number and route any track assigned to this output only, so they will be routed directly from within the DSP card instead, thus bypassing any plugin on the track. It is also important to understand that LLM works by muting the record armed track output only during recording, so you will still be able to perform punch in and overdubs.
When using Pro Tools HD 10 or equivalent with third-party interfaces, LLM will be limited to a single-stereo output that cannot be changed: output 1 and 2. You will also need to make sure that your soundcard has an onboard virtual mixer to route the desired signal to output 1 and 2, or you will hear nothing during recording as LLM mutes software monitoring while recording.
With Pro Tools HD 11, the same rules apply when using LLM; the only change is that we have another option with the dual-buffer technology. If you manage to set it low enough, it is the most convenient way of performing any overdub and you will not need to use LLM.
Now, when it comes to creating a cue mix with LLM engaged in either HD 11 or 10, the traditional way of creating an internal aux send mix to a separate output is not possible anymore unless we own an HD Native card, so we can select a different LLM output other than our mix output. When using third-party interfaces, we have no choice but to create the cue mix from within the soundcard mixer.
Pro Tools is a complex system with many different versions of the hardware and software and Version 10, offering a dual-installation method, will be the last version to support legacy TDM plugins alongside RTAS. Pro Tools 11 abandoned all legacy code and technologies to create a much more accurate signal path. To recapitulate what we have seen so far, here is a comparative table: