Designing the Tensor

2015

In April, 2015, the Raster went into production and I started working on our next pedal. Several years earlier, someone suggested that we make something like the tape stop effect on the old Digitech XP-300 Space Station. I had been kicking around the idea for a few years, but the reality of running a small company is that most of your time is spent on bookkeeping, ordering parts, and managing production. The Raster was delayed three months because of powder coating issues with the first batch of enclosures.

Back in 2010, I chose the relatively new Spin FV-1 DSP chip for the Particle. The Context, Bitmap, and Raster also use the FV-1, which is fun to program and pushes you to think about effects in a creative way. For the new pedal, I switched to a faster processor that could handle more advanced signal analysis and allowed me to reuse DSP code that I had written over the previous 20 years.

During the summer of 2015 I started designing a new hardware platform, set up a development environment, and wrote code to bootstrap the processor and handle basic functions like reading potentiometers and getting audio in and out. I was also working on another product using the FV-1. Most of the fall and winter were spent getting our new UV LED printer up and running and dealing with production emergencies, but I made some incremental progress on the Tensor. The first prototype had 1.2 seconds of sample memory.

The Tensor was initially focused on doing tape stop, slowdown, and reverse effects in real time. In September, 2015, Denver Dalley did a video with Reverb showing some of his innovative pedal tricks. For the stuttering effect on "City on the Hill", he used the hold function on the Boss DD-6 to repeat part of a note, then a Whammy to pitch shift it. That was the inspiration for the Tensor's Hold function and the Pitch knob. When grabbing part of a note, it is nice to have forward, reverse, or alternating (forward/reverse) loop playback like a sampler.

If can control tape speed and pitch, the obvious next step is time stretching. Like the Roland V-Synth, but in real-time, instead of sample playback. The Tensor constantly analyzes what you play so that you can manipulate direction, speed, pitch, and time stretch in any combination while minimizing latency. I spent most of the winter and spring of 2016 writing the analysis code and testing with a wide variety of instruments, signals, and music. I also did several more hardware prototypes for different size enclosures. I also upgraded to a faster processor and doubled the sampling memory to 2.4 seconds. At the end of 2015, the Tensor code was still a generic development board with an add-on board for potentiometers and audio input/output.

2016

In January 2016, I got the first hardware prototype up and running and it started to look like a pedal. There was still a lot of work to do, so I decided not to send the prototype to NAMM. We also shelved another product that was nearly ready to go.

The Tensor is always listening, so you can go straight from bypass to rewind or loop what you just played. If you rewind beyond the memory limit, it does a tape stop effect. You can go directly from bypass to hold. Holding down the right footswitch temporarily breaks from real-time and lets it drift. The goal was to allow intuitive and flexible sound manipulation with independent control over direction, time, and pitch, maintaining sound quality and playability so that you can focus on your playing.

The rest of 2016 was consumed by switching some suppliers and automated more of our manufacturing to increase production. In between, I spent time fine-tuning pitch shifting and time stretching.

NAMM 2017

NAMM never comes at a convenient time. In early January, 2017, I grabbed a blank Raster blem enclosure and laid out the control labels. Sylvie suggested covering the prototype in car camoflauge, which car companies use to obscure details of new car designs when they road test them (a common sight on roads here in Detroit). Up to this point it was called "Time Mechanic", after the Jeff Mills records. That described the pedal well, but was long and didn't fit with our other pedal names. A week before NAMM, I walked to a restaurant near our workshop to have lunch and think of a name and settled on "Tensor". Sylvie added it to her graphics and printed the prototype.

We had a lot of artist demos in our booth, which allowed me to do usability testing and get feedback. The default knob positions caused some confusion. People thought having time at noon meant no time stretching, but in fact the knob went from 0 to 4x time stretching with 2x at noon. The obvious thing to do was make noon no stretch, and add time compression to the right. To make pitch jumps easier, I added a momentary/latching switch for the bypass mode. Another momentary/latching switch on the hold footswitch lets you set it up for either stutters or longer loops.

2017

People loved the graphics, so we kept the idea and expanded it to add a bit of color and more depth. Sylvie meticulously tweaked, experimented, and fine tuned the graphics. We print them in house, so we can try out new ideas quickly. Sylvie pushed the process and came up with a design that catches the light in interesting ways.

A faster processor became available, so I upgraded again and doubled the sampling memory to 4.8 seconds. I changed to a smaller enclosure, shorter than the Raster and as narrow as possible with all jacks on top.

By this time the hold modes were fully implemented:

  • “REC” (record) grabs little fragments or loops.
  • “OVR” (overdub) does sound on sound. It records at the set playback speed/time/pitch, and with randomization it drops little fragments of sound at different spots in the loop.
  • NXT (“next”) mode was inspired by techniques Henry Kaiser pioneered years ago, using rack mount digital delays to play live reverse solos. Once you wrap your head around playing ahead of the notes you want to hear, it allows you to play solos that seem impossible to play live.

During the summer I implemented USB firmware updates and continued fine-tuning with different instruments and inputs. We showed it to several artists who came through Detroit on tour.

In early fall I implemented randomization. It reacts to the other knobs settings and changes the behavior of the randomness as you crank it up. I spent a lot of time fine-tuning the randomization to enable occasional glitches, stutters, repeats, random pitch jumps, and beat slicing. The rest of the year was spent making sure every combination of knob settings worked lived and looping with all kinds of signals. Almost all of the music I listened to in 2017 was mangled by the Tensor.

In October we ordered parts for the first production run. Everything was scheduled for an end of year release, so we did a preorder on Black Friday.

DECEMBER 2017

In early December we got the first 20 Tensor circuit boards (this was the 5th revision of the board). We spent a lot of time testing, under a lot of different conditions, and in some situations the Tensor was just a little noisy. I spent a week testing it and experimenting with software fixes, which helped but not as much as I wanted. All of the other parts were sitting on the shelf, and we could have built the first batch and fixed this in a later revision, but I didn’t want to have a noisy batch of Tensors out in the world. I changed to a higher-performance codec with its own power regulator, which also needed more support components. It costs us a lot more to build (but we didn't raise the price - benefit of being an engineer with no boss). I made the decision to change the design on December 19th, chose a new codec on the 20th, finished up the design on the 25th. It took a while to get quotes over the holiday break, but finally placed the order on the 29th with 3-day turn time on the boards and assembly, and got the boards January 10th. While waiting, I finished up our new packaging.

2018

Changing the firmware only took a couple of hours. Unfortunately, the codec change meant that the preorders would be delayed (everyone was very understanding) and I shelved some stuff we were going to show at NAMM. The other downside is that the current consumption went from 165 mA to 210 mA, which eliminated some power supplies. But if there is any place that’s worth investing some money and current, it is the converters and op amps (the Tensor uses Burr-Brown op amps).

The production circuit boards were ready on February 5th. Sylvie and Randy started building and testing pedals, and we shipped the first Tensors on February 8th, 2018. I started working on the Particle 2 on February 14th.